vpx/vp8/encoder/bitstream.c

2135 lines
66 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include "vp8/common/header.h"
2010-05-18 17:58:33 +02:00
#include "encodemv.h"
#include "vp8/common/entropymode.h"
#include "vp8/common/findnearmv.h"
2010-05-18 17:58:33 +02:00
#include "mcomp.h"
#include "vp8/common/systemdependent.h"
2010-05-18 17:58:33 +02:00
#include <assert.h>
#include <stdio.h>
#include <limits.h>
#include "vp8/common/pragmas.h"
#include "vpx/vpx_encoder.h"
2010-05-18 17:58:33 +02:00
#include "vpx_mem/vpx_mem.h"
#include "bitstream.h"
#include "defaultcoefcounts.h"
2011-10-05 12:26:00 +02:00
#include "vp8/common/seg_common.h"
#include "vp8/common/pred_common.h"
2011-10-05 12:26:00 +02:00
2010-05-18 17:58:33 +02:00
#if defined(SECTIONBITS_OUTPUT)
unsigned __int64 Sectionbits[500];
#endif
#ifdef ENTROPY_STATS
int intra_mode_stats[10][10][10];
static unsigned int tree_update_hist [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES] [2];
static unsigned int tree_update_hist_8x8 [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES] [2];
2010-05-18 17:58:33 +02:00
extern unsigned int active_section;
#endif
#ifdef MODE_STATS
int count_mb_seg[4] = { 0, 0, 0, 0 };
#endif
static void update_mode(
vp8_writer *const w,
int n,
vp8_token tok [/* n */],
vp8_tree tree,
vp8_prob Pnew [/* n-1 */],
vp8_prob Pcur [/* n-1 */],
unsigned int bct [/* n-1 */] [2],
const unsigned int num_events[/* n */]
)
{
unsigned int new_b = 0, old_b = 0;
int i = 0;
vp8_tree_probs_from_distribution(
n--, tok, tree,
Pnew, bct, num_events,
256, 1
);
do
{
new_b += vp8_cost_branch(bct[i], Pnew[i]);
old_b += vp8_cost_branch(bct[i], Pcur[i]);
}
while (++i < n);
if (new_b + (n << 8) < old_b)
{
int i = 0;
vp8_write_bit(w, 1);
do
{
const vp8_prob p = Pnew[i];
vp8_write_literal(w, Pcur[i] = p ? p : 1, 8);
}
while (++i < n);
}
else
vp8_write_bit(w, 0);
}
static void update_mbintra_mode_probs(VP8_COMP *cpi)
{
VP8_COMMON *const x = & cpi->common;
vp8_writer *const w = & cpi->bc;
{
vp8_prob Pnew [VP8_YMODES-1];
unsigned int bct [VP8_YMODES-1] [2];
update_mode(
w, VP8_YMODES, vp8_ymode_encodings, vp8_ymode_tree,
Pnew, x->fc.ymode_prob, bct, (unsigned int *)cpi->ymode_count
);
}
{
#if CONFIG_UVINTRA
//vp8_write_bit(w, 0);
#else
2010-05-18 17:58:33 +02:00
vp8_prob Pnew [VP8_UV_MODES-1];
unsigned int bct [VP8_UV_MODES-1] [2];
update_mode(
w, VP8_UV_MODES, vp8_uv_mode_encodings, vp8_uv_mode_tree,
Pnew, x->fc.uv_mode_prob, bct, (unsigned int *)cpi->uv_mode_count
);
#endif
2010-05-18 17:58:33 +02:00
}
}
static void write_ymode(vp8_writer *bc, int m, const vp8_prob *p)
{
vp8_write_token(bc, vp8_ymode_tree, p, vp8_ymode_encodings + m);
}
static void kfwrite_ymode(vp8_writer *bc, int m, const vp8_prob *p)
{
vp8_write_token(bc, vp8_kf_ymode_tree, p, vp8_kf_ymode_encodings + m);
}
static void write_i8x8_mode(vp8_writer *bc, int m, const vp8_prob *p)
{
vp8_write_token(bc,vp8_i8x8_mode_tree, p, vp8_i8x8_mode_encodings + m);
}
2010-05-18 17:58:33 +02:00
static void write_uv_mode(vp8_writer *bc, int m, const vp8_prob *p)
{
vp8_write_token(bc, vp8_uv_mode_tree, p, vp8_uv_mode_encodings + m);
}
static void write_bmode(vp8_writer *bc, int m, const vp8_prob *p)
{
vp8_write_token(bc, vp8_bmode_tree, p, vp8_bmode_encodings + m);
}
static void write_split(vp8_writer *bc, int x)
{
vp8_write_token(
bc, vp8_mbsplit_tree, vp8_mbsplit_probs, vp8_mbsplit_encodings + x
);
}
static void pack_tokens_c(vp8_writer *w, const TOKENEXTRA *p, int xcount)
{
const TOKENEXTRA *const stop = p + xcount;
unsigned int split;
unsigned int shift;
int count = w->count;
unsigned int range = w->range;
unsigned int lowvalue = w->lowvalue;
while (p < stop)
{
const int t = p->Token;
vp8_token *const a = vp8_coef_encodings + t;
const vp8_extra_bit_struct *const b = vp8_extra_bits + t;
int i = 0;
const unsigned char *pp = p->context_tree;
int v = a->value;
int n = a->Len;
if (p->skip_eob_node)
{
n--;
i = 2;
}
do
{
const int bb = (v >> --n) & 1;
split = 1 + (((range - 1) * pp[i>>1]) >> 8);
i = vp8_coef_tree[i+bb];
if (bb)
{
lowvalue += split;
range = range - split;
}
else
{
range = split;
}
shift = vp8_norm[range];
2010-05-18 17:58:33 +02:00
range <<= shift;
count += shift;
if (count >= 0)
{
int offset = shift - count;
if ((lowvalue << (offset - 1)) & 0x80000000)
{
int x = w->pos - 1;
while (x >= 0 && w->buffer[x] == 0xff)
{
w->buffer[x] = (unsigned char)0;
x--;
}
w->buffer[x] += 1;
}
w->buffer[w->pos++] = (lowvalue >> (24 - offset));
lowvalue <<= offset;
shift = count;
lowvalue &= 0xffffff;
count -= 8 ;
}
lowvalue <<= shift;
}
while (n);
if (b->base_val)
{
const int e = p->Extra, L = b->Len;
if (L)
{
const unsigned char *pp = b->prob;
int v = e >> 1;
int n = L; /* number of bits in v, assumed nonzero */
int i = 0;
do
{
const int bb = (v >> --n) & 1;
split = 1 + (((range - 1) * pp[i>>1]) >> 8);
i = b->tree[i+bb];
if (bb)
{
lowvalue += split;
range = range - split;
}
else
{
range = split;
}
shift = vp8_norm[range];
2010-05-18 17:58:33 +02:00
range <<= shift;
count += shift;
if (count >= 0)
{
int offset = shift - count;
if ((lowvalue << (offset - 1)) & 0x80000000)
{
int x = w->pos - 1;
while (x >= 0 && w->buffer[x] == 0xff)
{
w->buffer[x] = (unsigned char)0;
x--;
}
w->buffer[x] += 1;
}
w->buffer[w->pos++] = (lowvalue >> (24 - offset));
lowvalue <<= offset;
shift = count;
lowvalue &= 0xffffff;
count -= 8 ;
}
lowvalue <<= shift;
}
while (n);
}
{
split = (range + 1) >> 1;
if (e & 1)
{
lowvalue += split;
range = range - split;
}
else
{
range = split;
}
range <<= 1;
if ((lowvalue & 0x80000000))
{
int x = w->pos - 1;
while (x >= 0 && w->buffer[x] == 0xff)
{
w->buffer[x] = (unsigned char)0;
x--;
}
w->buffer[x] += 1;
}
lowvalue <<= 1;
if (!++count)
{
count = -8;
w->buffer[w->pos++] = (lowvalue >> 24);
lowvalue &= 0xffffff;
}
}
}
++p;
}
w->count = count;
w->lowvalue = lowvalue;
w->range = range;
}
static void write_partition_size(unsigned char *cx_data, int size)
{
signed char csize;
csize = size & 0xff;
*cx_data = csize;
csize = (size >> 8) & 0xff;
*(cx_data + 1) = csize;
csize = (size >> 16) & 0xff;
*(cx_data + 2) = csize;
}
static void write_mv_ref
(
vp8_writer *w, MB_PREDICTION_MODE m, const vp8_prob *p
)
{
#if CONFIG_DEBUG
2010-05-18 17:58:33 +02:00
assert(NEARESTMV <= m && m <= SPLITMV);
#endif
vp8_write_token(w, vp8_mv_ref_tree, p,
vp8_mv_ref_encoding_array - NEARESTMV + m);
2010-05-18 17:58:33 +02:00
}
static void write_sub_mv_ref
(
vp8_writer *w, B_PREDICTION_MODE m, const vp8_prob *p
)
{
#if CONFIG_DEBUG
2010-05-18 17:58:33 +02:00
assert(LEFT4X4 <= m && m <= NEW4X4);
#endif
vp8_write_token(w, vp8_sub_mv_ref_tree, p,
vp8_sub_mv_ref_encoding_array - LEFT4X4 + m);
2010-05-18 17:58:33 +02:00
}
static void write_mv
(
vp8_writer *w, const MV *mv, const int_mv *ref, const MV_CONTEXT *mvc
2010-05-18 17:58:33 +02:00
)
{
MV e;
e.row = mv->row - ref->as_mv.row;
e.col = mv->col - ref->as_mv.col;
2010-05-18 17:58:33 +02:00
vp8_encode_motion_vector(w, &e, mvc);
}
#if CONFIG_HIGH_PRECISION_MV
static void write_mv_hp
(
vp8_writer *w, const MV *mv, const int_mv *ref, const MV_CONTEXT_HP *mvc
)
{
MV e;
e.row = mv->row - ref->as_mv.row;
e.col = mv->col - ref->as_mv.col;
vp8_encode_motion_vector_hp(w, &e, mvc);
}
#endif
// This function writes the current macro block's segnment id to the bitstream
// It should only be called if a segment map update is indicated.
static void write_mb_segid(vp8_writer *w,
const MB_MODE_INFO *mi, const MACROBLOCKD *x)
2010-05-18 17:58:33 +02:00
{
// Encode the MB segment id.
if (x->segmentation_enabled && x->update_mb_segmentation_map)
{
switch (mi->segment_id)
{
case 0:
vp8_write(w, 0, x->mb_segment_tree_probs[0]);
vp8_write(w, 0, x->mb_segment_tree_probs[1]);
break;
case 1:
vp8_write(w, 0, x->mb_segment_tree_probs[0]);
vp8_write(w, 1, x->mb_segment_tree_probs[1]);
break;
case 2:
vp8_write(w, 1, x->mb_segment_tree_probs[0]);
vp8_write(w, 0, x->mb_segment_tree_probs[2]);
break;
case 3:
vp8_write(w, 1, x->mb_segment_tree_probs[0]);
vp8_write(w, 1, x->mb_segment_tree_probs[2]);
break;
// TRAP.. This should not happen
default:
vp8_write(w, 0, x->mb_segment_tree_probs[0]);
vp8_write(w, 0, x->mb_segment_tree_probs[1]);
break;
}
}
}
// This function encodes the reference frame
static void encode_ref_frame( vp8_writer *const w,
VP8_COMMON *const cm,
MACROBLOCKD *xd,
int segment_id,
MV_REFERENCE_FRAME rf )
{
int seg_ref_active;
int seg_ref_count = 0;
seg_ref_active = segfeature_active( xd,
segment_id,
SEG_LVL_REF_FRAME );
if ( seg_ref_active )
{
seg_ref_count = check_segref( xd, segment_id, INTRA_FRAME ) +
check_segref( xd, segment_id, LAST_FRAME ) +
check_segref( xd, segment_id, GOLDEN_FRAME ) +
check_segref( xd, segment_id, ALTREF_FRAME );
}
// If segment level coding of this signal is disabled...
// or the segment allows multiple reference frame options
if ( !seg_ref_active || (seg_ref_count > 1) )
{
// Values used in prediction model coding
unsigned char prediction_flag;
vp8_prob pred_prob;
MV_REFERENCE_FRAME pred_rf;
// Get the context probability the prediction flag
pred_prob = get_pred_prob( cm, xd, PRED_REF );
// Get the predicted value.
pred_rf = get_pred_ref( cm, xd );
// Did the chosen reference frame match its predicted value.
prediction_flag =
( xd->mode_info_context->mbmi.ref_frame == pred_rf );
set_pred_flag( xd, PRED_REF, prediction_flag );
vp8_write( w, prediction_flag, pred_prob );
// If not predicted correctly then code value explicitly
if ( !prediction_flag )
{
vp8_prob mod_refprobs[PREDICTION_PROBS];
vpx_memcpy( mod_refprobs,
cm->mod_refprobs[pred_rf], sizeof(mod_refprobs) );
// If segment coding enabled blank out options that cant occur by
// setting the branch probability to 0.
if ( seg_ref_active )
{
mod_refprobs[INTRA_FRAME] *=
check_segref( xd, segment_id, INTRA_FRAME );
mod_refprobs[LAST_FRAME] *=
check_segref( xd, segment_id, LAST_FRAME );
mod_refprobs[GOLDEN_FRAME] *=
( check_segref( xd, segment_id, GOLDEN_FRAME ) *
check_segref( xd, segment_id, ALTREF_FRAME ) );
}
if ( mod_refprobs[0] )
{
vp8_write(w, (rf != INTRA_FRAME), mod_refprobs[0] );
}
// Inter coded
if (rf != INTRA_FRAME)
{
if ( mod_refprobs[1] )
{
vp8_write(w, (rf != LAST_FRAME), mod_refprobs[1] );
}
if (rf != LAST_FRAME)
{
if ( mod_refprobs[2] )
{
vp8_write(w, (rf != GOLDEN_FRAME), mod_refprobs[2] );
}
}
}
}
}
// if using the prediction mdoel we have nothing further to do because
// the reference frame is fully coded by the segment
}
2010-05-18 17:58:33 +02:00
// Update the probabilities used to encode reference frame data
static void update_ref_probs( VP8_COMP *const cpi )
{
VP8_COMMON *const cm = & cpi->common;
const int *const rfct = cpi->count_mb_ref_frame_usage;
const int rf_intra = rfct[INTRA_FRAME];
const int rf_inter = rfct[LAST_FRAME] +
rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME];
cm->prob_intra_coded = (rf_intra + rf_inter)
? rf_intra * 255 / (rf_intra + rf_inter) : 1;
if (!cm->prob_intra_coded)
cm->prob_intra_coded = 1;
cm->prob_last_coded = rf_inter ? (rfct[LAST_FRAME] * 255) / rf_inter : 128;
if (!cm->prob_last_coded)
cm->prob_last_coded = 1;
cm->prob_gf_coded = (rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME])
? (rfct[GOLDEN_FRAME] * 255) /
(rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME]) : 128;
if (!cm->prob_gf_coded)
cm->prob_gf_coded = 1;
// Compute a modified set of probabilities to use when prediction of the
// reference frame fails
compute_mod_refprobs( cm );
}
2010-05-18 17:58:33 +02:00
static void pack_inter_mode_mvs(VP8_COMP *const cpi)
{
VP8_COMMON *const pc = & cpi->common;
vp8_writer *const w = & cpi->bc;
const MV_CONTEXT *mvc = pc->fc.mvc;
#if CONFIG_HIGH_PRECISION_MV
const MV_CONTEXT_HP *mvc_hp = pc->fc.mvc_hp;
#endif
MACROBLOCKD *xd = &cpi->mb.e_mbd;
int i;
Further work on Segmentation Experiment: This check in includes quite a lot of clean up and refactoring. Most of the analysis and set up for the different coding options for the segment map (currently simple distribution based coding or temporaly predicted coding), has been moved to one location (the function choose_segmap_coding_method() in segmenation.c). This code was previously scattered around in various locations making integration with other experiments and modification / debug more difficult. Currently the functionality is as it was with the exception that the prediction probabilities are now only transmitted when the temporal prediction mode is selected. There is still quite a bit more clean up work that will be possible when the #ifdef is removed. Also at that time I may rename and alter the sense of macroblock based variable "segment_flag" which indicates (1 that the segmnet id is not predicted vs 0 that it is predicted). I also intend to experiment with a spatial prediction mode that can be used when coding a key frame segment map or in cases where temporal prediction does not work well but there is spatial correlation. In a later check in when the ifdefs have gone I may also move the call to choose_segmap_coding_method() to just before where the bitsream is packed (currently it is in vp8_encode_frame()) to further reduce the possibility of clashes with other experiments and prevent it being called on each itteration of the recode loop. Change-Id: I3d4aba2a2826ec21f367678d5b07c1d1c36db168
2011-11-15 12:13:33 +01:00
int pred_context;
2010-05-18 17:58:33 +02:00
MODE_INFO *m = pc->mi;
MODE_INFO *prev_m = pc->prev_mi;
2010-05-18 17:58:33 +02:00
const int mis = pc->mode_info_stride;
int mb_row = -1;
int prob_skip_false = 0;
// Values used in prediction model coding
vp8_prob pred_prob;
unsigned char prediction_flag;
cpi->mb.partition_info = cpi->mb.pi;
// Update the probabilities used to encode reference frame data
update_ref_probs( cpi );
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
active_section = 1;
#endif
if (pc->mb_no_coeff_skip)
{
// Divide by 0 check. 0 case possible with segment features
if ( (cpi->skip_false_count + cpi->skip_true_count) )
{
prob_skip_false = cpi->skip_false_count * 256 /
(cpi->skip_false_count + cpi->skip_true_count);
2010-05-18 17:58:33 +02:00
if (prob_skip_false <= 1)
prob_skip_false = 1;
2010-05-18 17:58:33 +02:00
if (prob_skip_false > 255)
prob_skip_false = 255;
}
else
2010-05-18 17:58:33 +02:00
prob_skip_false = 255;
cpi->prob_skip_false = prob_skip_false;
vp8_write_literal(w, prob_skip_false, 8);
}
vp8_write_literal(w, pc->prob_intra_coded, 8);
vp8_write_literal(w, pc->prob_last_coded, 8);
vp8_write_literal(w, pc->prob_gf_coded, 8);
2010-05-18 17:58:33 +02:00
if (cpi->common.comp_pred_mode == HYBRID_PREDICTION)
{
vp8_write(w, 1, 128);
vp8_write(w, 1, 128);
for (i = 0; i < COMP_PRED_CONTEXTS; i++)
{
if (cpi->single_pred_count[i] + cpi->comp_pred_count[i])
{
pc->prob_comppred[i] = cpi->single_pred_count[i] * 255 /
(cpi->single_pred_count[i] + cpi->comp_pred_count[i]);
if (pc->prob_comppred[i] < 1)
pc->prob_comppred[i] = 1;
}
else
{
pc->prob_comppred[i] = 128;
}
vp8_write_literal(w, pc->prob_comppred[i], 8);
}
}
else if (cpi->common.comp_pred_mode == SINGLE_PREDICTION_ONLY)
{
vp8_write(w, 0, 128);
}
else /* compound prediction only */
{
vp8_write(w, 1, 128);
vp8_write(w, 0, 128);
}
2010-05-18 17:58:33 +02:00
update_mbintra_mode_probs(cpi);
#if CONFIG_HIGH_PRECISION_MV
if (xd->allow_high_precision_mv)
vp8_write_mvprobs_hp(cpi);
else
#endif
2010-05-18 17:58:33 +02:00
vp8_write_mvprobs(cpi);
2010-05-18 17:58:33 +02:00
while (++mb_row < pc->mb_rows)
{
int mb_col = -1;
while (++mb_col < pc->mb_cols)
{
const MB_MODE_INFO *const mi = & m->mbmi;
const MV_REFERENCE_FRAME rf = mi->ref_frame;
const MB_PREDICTION_MODE mode = mi->mode;
2011-10-05 12:26:00 +02:00
const int segment_id = mi->segment_id;
2010-05-18 17:58:33 +02:00
// Distance of Mb to the various image edges.
// These specified to 8th pel as they are always compared to MV values that are in 1/8th pel units
xd->mb_to_left_edge = -((mb_col * 16) << 3);
xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3;
xd->mb_to_top_edge = -((mb_row * 16)) << 3;
xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3;
// Make sure the MacroBlockD mode info pointer is set correctly
xd->mode_info_context = m;
xd->prev_mode_info_context = prev_m;
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
active_section = 9;
#endif
if (cpi->mb.e_mbd.update_mb_segmentation_map)
{
// Is temporal coding of the segment map enabled
if (pc->temporal_update)
{
prediction_flag =
get_pred_flag( xd, PRED_SEG_ID );
pred_prob =
get_pred_prob( pc, xd, PRED_SEG_ID);
// Code the segment id prediction flag for this mb
vp8_write( w, prediction_flag, pred_prob );
// If the mbs segment id was not predicted code explicitly
if (!prediction_flag)
write_mb_segid(w, mi, &cpi->mb.e_mbd);
}
else
{
// Normal undpredicted coding
write_mb_segid(w, mi, &cpi->mb.e_mbd);
}
}
2010-05-18 17:58:33 +02:00
2011-10-05 12:26:00 +02:00
if ( pc->mb_no_coeff_skip &&
( !segfeature_active( xd, segment_id, SEG_LVL_EOB ) ||
( get_segdata( xd, segment_id, SEG_LVL_EOB ) != 0 ) ) )
2011-10-05 12:26:00 +02:00
{
vp8_encode_bool(w, mi->mb_skip_coeff, prob_skip_false);
}
2010-05-18 17:58:33 +02:00
// Encode the reference frame.
encode_ref_frame( w, pc, xd,
segment_id, rf );
2010-05-18 17:58:33 +02:00
if (rf == INTRA_FRAME)
{
#ifdef ENTROPY_STATS
2010-05-18 17:58:33 +02:00
active_section = 6;
#endif
if ( !segfeature_active( xd, segment_id, SEG_LVL_MODE ) )
write_ymode(w, mode, pc->fc.ymode_prob);
2010-05-18 17:58:33 +02:00
if (mode == B_PRED)
{
int j = 0;
#if CONFIG_COMP_INTRA_PRED
int uses_second = m->bmi[0].as_mode.second != (B_PREDICTION_MODE) (B_DC_PRED - 1);
vp8_write(w, uses_second, 128);
#endif
do {
#if CONFIG_COMP_INTRA_PRED
B_PREDICTION_MODE mode2 = m->bmi[j].as_mode.second;
#endif
write_bmode(w, m->bmi[j].as_mode.first, pc->fc.bmode_prob);
#if CONFIG_COMP_INTRA_PRED
if (uses_second)
{
write_bmode(w, mode2, pc->fc.bmode_prob);
}
#endif
} while (++j < 16);
2010-05-18 17:58:33 +02:00
}
if(mode == I8X8_PRED)
{
write_i8x8_mode(w, m->bmi[0].as_mode.first, pc->i8x8_mode_prob);
write_i8x8_mode(w, m->bmi[2].as_mode.first, pc->i8x8_mode_prob);
write_i8x8_mode(w, m->bmi[8].as_mode.first, pc->i8x8_mode_prob);
write_i8x8_mode(w, m->bmi[10].as_mode.first, pc->i8x8_mode_prob);
}
else
{
#if CONFIG_UVINTRA
write_uv_mode(w, mi->uv_mode, pc->fc.uv_mode_prob[mode]);
#ifdef MODE_STATS
if(mode!=B_PRED)
++cpi->y_uv_mode_count[mode][mi->uv_mode];
#endif
#else
write_uv_mode(w, mi->uv_mode, pc->fc.uv_mode_prob);
#endif /*CONFIG_UVINTRA*/
}
2010-05-18 17:58:33 +02:00
}
else
2010-05-18 17:58:33 +02:00
{
int_mv best_mv;
int ct[4];
2010-05-18 17:58:33 +02:00
vp8_prob mv_ref_p [VP8_MVREFS-1];
{
int_mv n1, n2;
2010-05-18 17:58:33 +02:00
vp8_find_near_mvs(xd, m,
prev_m,
&n1, &n2, &best_mv, ct, rf, cpi->common.ref_frame_sign_bias);
vp8_mv_ref_probs(&cpi->common, mv_ref_p, ct);
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
accum_mv_refs(mode, ct);
#endif
}
#ifdef ENTROPY_STATS
active_section = 3;
#endif
// Is the segment coding of mode enabled
2011-10-05 12:26:00 +02:00
if ( !segfeature_active( xd, segment_id, SEG_LVL_MODE ) )
2010-05-18 17:58:33 +02:00
{
write_mv_ref(w, mode, mv_ref_p);
vp8_accum_mv_refs(&cpi->common, mode, ct);
}
{
switch (mode) /* new, split require MVs */
{
case NEWMV:
#ifdef ENTROPY_STATS
active_section = 5;
#endif
2010-05-18 17:58:33 +02:00
#if CONFIG_HIGH_PRECISION_MV
if (xd->allow_high_precision_mv)
write_mv_hp(w, &mi->mv.as_mv, &best_mv, mvc_hp);
else
#endif
write_mv(w, &mi->mv.as_mv, &best_mv, mvc);
if (cpi->common.comp_pred_mode == HYBRID_PREDICTION)
{
vp8_write(w, mi->second_ref_frame != INTRA_FRAME,
get_pred_prob( pc, xd, PRED_COMP ) );
}
if (mi->second_ref_frame)
{
const int second_rf = mi->second_ref_frame;
int_mv n1, n2;
int ct[4];
vp8_find_near_mvs(xd, m,
prev_m,
&n1, &n2, &best_mv,
ct, second_rf,
cpi->common.ref_frame_sign_bias);
#if CONFIG_HIGH_PRECISION_MV
if (xd->allow_high_precision_mv)
write_mv_hp(w, &mi->second_mv.as_mv, &best_mv, mvc_hp);
else
#endif
write_mv(w, &mi->second_mv.as_mv, &best_mv, mvc);
}
break;
case SPLITMV:
2010-05-18 17:58:33 +02:00
{
int j = 0;
2010-05-18 17:58:33 +02:00
#ifdef MODE_STATS
++count_mb_seg [mi->partitioning];
#endif
2010-05-18 17:58:33 +02:00
write_split(w, mi->partitioning);
do
2010-05-18 17:58:33 +02:00
{
B_PREDICTION_MODE blockmode;
int_mv blockmv;
const int *const L = vp8_mbsplits [mi->partitioning];
int k = -1; /* first block in subset j */
int mv_contz;
int_mv leftmv, abovemv;
blockmode = cpi->mb.partition_info->bmi[j].mode;
blockmv = cpi->mb.partition_info->bmi[j].mv;
#if CONFIG_DEBUG
while (j != L[++k])
if (k >= 16)
assert(0);
#else
while (j != L[++k]);
#endif
leftmv.as_int = left_block_mv(m, k);
abovemv.as_int = above_block_mv(m, k, mis);
mv_contz = vp8_mv_cont(&leftmv, &abovemv);
write_sub_mv_ref(w, blockmode, vp8_sub_mv_ref_prob2 [mv_contz]);
if (blockmode == NEW4X4)
{
#ifdef ENTROPY_STATS
active_section = 11;
#endif
#if CONFIG_HIGH_PRECISION_MV
if (xd->allow_high_precision_mv)
write_mv_hp(w, &blockmv.as_mv, &best_mv, (const MV_CONTEXT_HP *) mvc_hp);
else
#endif
write_mv(w, &blockmv.as_mv, &best_mv, (const MV_CONTEXT *) mvc);
}
2010-05-18 17:58:33 +02:00
}
while (++j < cpi->mb.partition_info->count);
2010-05-18 17:58:33 +02:00
}
break;
default:
if (cpi->common.comp_pred_mode == HYBRID_PREDICTION)
{
vp8_write(w, mi->second_ref_frame != INTRA_FRAME,
get_pred_prob( pc, xd, PRED_COMP ) );
}
break;
}
2010-05-18 17:58:33 +02:00
}
}
++m;
++prev_m;
assert((prev_m-cpi->common.prev_mip)==(m-cpi->common.mip));
assert((prev_m-cpi->common.prev_mi)==(m-cpi->common.mi));
cpi->mb.partition_info++;
2010-05-18 17:58:33 +02:00
}
++m; /* skip L prediction border */
++prev_m;
cpi->mb.partition_info++;
2010-05-18 17:58:33 +02:00
}
}
2010-05-18 17:58:33 +02:00
static void write_kfmodes(VP8_COMP *cpi)
{
vp8_writer *const bc = & cpi->bc;
const VP8_COMMON *const c = & cpi->common;
/* const */
MODE_INFO *m = c->mi;
int mb_row = -1;
int prob_skip_false = 0;
2011-10-05 12:26:00 +02:00
MACROBLOCKD *xd = &cpi->mb.e_mbd;
2010-05-18 17:58:33 +02:00
if (c->mb_no_coeff_skip)
{
// Divide by 0 check. 0 case possible with segment features
if ( (cpi->skip_false_count + cpi->skip_true_count) )
{
prob_skip_false = cpi->skip_false_count * 256 /
(cpi->skip_false_count + cpi->skip_true_count);
2010-05-18 17:58:33 +02:00
if (prob_skip_false <= 1)
prob_skip_false = 1;
2010-05-18 17:58:33 +02:00
if (prob_skip_false > 255)
prob_skip_false = 255;
}
else
2010-05-18 17:58:33 +02:00
prob_skip_false = 255;
cpi->prob_skip_false = prob_skip_false;
vp8_write_literal(bc, prob_skip_false, 8);
}
#if CONFIG_QIMODE
if(!c->kf_ymode_probs_update)
{
vp8_write_literal(bc, c->kf_ymode_probs_index, 3);
}
#endif
2010-05-18 17:58:33 +02:00
while (++mb_row < c->mb_rows)
{
int mb_col = -1;
while (++mb_col < c->mb_cols)
{
const int ym = m->mbmi.mode;
2011-10-05 12:26:00 +02:00
int segment_id = m->mbmi.segment_id;
2010-05-18 17:58:33 +02:00
if (cpi->mb.e_mbd.update_mb_segmentation_map)
{
Further work on Segmentation Experiment: This check in includes quite a lot of clean up and refactoring. Most of the analysis and set up for the different coding options for the segment map (currently simple distribution based coding or temporaly predicted coding), has been moved to one location (the function choose_segmap_coding_method() in segmenation.c). This code was previously scattered around in various locations making integration with other experiments and modification / debug more difficult. Currently the functionality is as it was with the exception that the prediction probabilities are now only transmitted when the temporal prediction mode is selected. There is still quite a bit more clean up work that will be possible when the #ifdef is removed. Also at that time I may rename and alter the sense of macroblock based variable "segment_flag" which indicates (1 that the segmnet id is not predicted vs 0 that it is predicted). I also intend to experiment with a spatial prediction mode that can be used when coding a key frame segment map or in cases where temporal prediction does not work well but there is spatial correlation. In a later check in when the ifdefs have gone I may also move the call to choose_segmap_coding_method() to just before where the bitsream is packed (currently it is in vp8_encode_frame()) to further reduce the possibility of clashes with other experiments and prevent it being called on each itteration of the recode loop. Change-Id: I3d4aba2a2826ec21f367678d5b07c1d1c36db168
2011-11-15 12:13:33 +01:00
write_mb_segid(bc, &m->mbmi, &cpi->mb.e_mbd);
}
2010-05-18 17:58:33 +02:00
2011-10-05 12:26:00 +02:00
if ( c->mb_no_coeff_skip &&
( !segfeature_active( xd, segment_id, SEG_LVL_EOB ) ||
(get_segdata( xd, segment_id, SEG_LVL_EOB ) != 0) ) )
2011-10-05 12:26:00 +02:00
{
2010-05-18 17:58:33 +02:00
vp8_encode_bool(bc, m->mbmi.mb_skip_coeff, prob_skip_false);
2011-10-05 12:26:00 +02:00
}
#if CONFIG_QIMODE
kfwrite_ymode(bc, ym, c->kf_ymode_prob[c->kf_ymode_probs_index]);
#else
2010-05-18 17:58:33 +02:00
kfwrite_ymode(bc, ym, c->kf_ymode_prob);
#endif
2010-05-18 17:58:33 +02:00
if (ym == B_PRED)
{
const int mis = c->mode_info_stride;
int i = 0;
#if CONFIG_COMP_INTRA_PRED
int uses_second = m->bmi[0].as_mode.second != (B_PREDICTION_MODE) (B_DC_PRED - 1);
vp8_write(bc, uses_second, 128);
#endif
2010-05-18 17:58:33 +02:00
do
{
const B_PREDICTION_MODE A = above_block_mode(m, i, mis);
const B_PREDICTION_MODE L = left_block_mode(m, i);
const int bm = m->bmi[i].as_mode.first;
#if CONFIG_COMP_INTRA_PRED
const int bm2 = m->bmi[i].as_mode.second;
#endif
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
++intra_mode_stats [A] [L] [bm];
#endif
write_bmode(bc, bm, c->kf_bmode_prob [A] [L]);
#if CONFIG_COMP_INTRA_PRED
if (uses_second)
{
write_bmode(bc, bm2, c->kf_bmode_prob [A] [L]);
}
#endif
2010-05-18 17:58:33 +02:00
}
while (++i < 16);
}
if(ym == I8X8_PRED)
{
write_i8x8_mode(bc, m->bmi[0].as_mode.first, c->i8x8_mode_prob);
write_i8x8_mode(bc, m->bmi[2].as_mode.first, c->i8x8_mode_prob);
write_i8x8_mode(bc, m->bmi[8].as_mode.first, c->i8x8_mode_prob);
write_i8x8_mode(bc, m->bmi[10].as_mode.first, c->i8x8_mode_prob);
m++;
}
else
#if CONFIG_UVINTRA
write_uv_mode(bc, (m++)->mbmi.uv_mode, c->kf_uv_mode_prob[ym]);
#else
write_uv_mode(bc, (m++)->mbmi.uv_mode, c->kf_uv_mode_prob);
#endif
2010-05-18 17:58:33 +02:00
}
//printf("\n");
2010-05-18 17:58:33 +02:00
m++; // skip L prediction border
}
}
/* This function is used for debugging probability trees. */
static void print_prob_tree(vp8_prob
coef_probs[BLOCK_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES])
2010-05-18 17:58:33 +02:00
{
/* print coef probability tree */
int i,j,k,l;
FILE* f = fopen("enc_tree_probs.txt", "a");
fprintf(f, "{\n");
for (i = 0; i < BLOCK_TYPES; i++)
{
fprintf(f, " {\n");
for (j = 0; j < COEF_BANDS; j++)
{
fprintf(f, " {\n");
for (k = 0; k < PREV_COEF_CONTEXTS; k++)
{
fprintf(f, " {");
for (l = 0; l < ENTROPY_NODES; l++)
{
fprintf(f, "%3u, ",
(unsigned int)(coef_probs [i][j][k][l]));
}
fprintf(f, " }\n");
}
fprintf(f, " }\n");
}
fprintf(f, " }\n");
}
fprintf(f, "}\n");
fclose(f);
}
static void sum_probs_over_prev_coef_context(
const unsigned int probs[PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS],
unsigned int* out)
{
int i, j;
for (i=0; i < MAX_ENTROPY_TOKENS; ++i)
{
for (j=0; j < PREV_COEF_CONTEXTS; ++j)
{
const int tmp = out[i];
out[i] += probs[j][i];
/* check for wrap */
if (out[i] < tmp)
out[i] = UINT_MAX;
}
}
}
static int prob_update_savings(const unsigned int *ct,
const vp8_prob oldp, const vp8_prob newp,
const vp8_prob upd)
{
const int old_b = vp8_cost_branch(ct, oldp);
const int new_b = vp8_cost_branch(ct, newp);
const int update_b = 8 +
((vp8_cost_one(upd) - vp8_cost_zero(upd)) >> 8);
return old_b - new_b - update_b;
}
static int default_coef_context_savings(VP8_COMP *cpi)
{
int savings = 0;
int i = 0;
do
{
int j = 0;
do
{
int k = 0;
do
{
/* at every context */
/* calc probs and branch cts for this frame only */
//vp8_prob new_p [ENTROPY_NODES];
//unsigned int branch_ct [ENTROPY_NODES] [2];
int t = 0; /* token/prob index */
vp8_tree_probs_from_distribution(
MAX_ENTROPY_TOKENS, vp8_coef_encodings, vp8_coef_tree,
cpi->frame_coef_probs [i][j][k],
cpi->frame_branch_ct [i][j][k],
cpi->coef_counts [i][j][k],
256, 1
);
do
{
const unsigned int *ct = cpi->frame_branch_ct [i][j][k][t];
const vp8_prob newp = cpi->frame_coef_probs [i][j][k][t];
const vp8_prob oldp = cpi->common.fc.coef_probs [i][j][k][t];
const vp8_prob upd = vp8_coef_update_probs [i][j][k][t];
const int s = prob_update_savings(ct, oldp, newp, upd);
if (s > 0)
{
savings += s;
}
}
while (++t < ENTROPY_NODES);
}
while (++k < PREV_COEF_CONTEXTS);
}
while (++j < COEF_BANDS);
}
while (++i < BLOCK_TYPES);
return savings;
}
int vp8_estimate_entropy_savings(VP8_COMP *cpi)
{
2010-05-18 17:58:33 +02:00
int savings = 0;
int i=0;
VP8_COMMON *const cm = & cpi->common;
2010-05-18 17:58:33 +02:00
const int *const rfct = cpi->count_mb_ref_frame_usage;
const int rf_intra = rfct[INTRA_FRAME];
const int rf_inter = rfct[LAST_FRAME] + rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME];
int new_intra, new_last, new_gf_alt, oldtotal, newtotal;
2010-05-18 17:58:33 +02:00
int ref_frame_cost[MAX_REF_FRAMES];
vp8_clear_system_state(); //__asm emms;
// Estimate reference frame cost savings.
// For now this is just based on projected overall frequency of
// each reference frame coded using an unpredicted coding tree.
2010-05-18 17:58:33 +02:00
if (cpi->common.frame_type != KEY_FRAME)
{
new_intra = (rf_intra + rf_inter)
? rf_intra * 255 / (rf_intra + rf_inter) : 1;
new_intra += !new_intra;
2010-05-18 17:58:33 +02:00
new_last = rf_inter ? (rfct[LAST_FRAME] * 255) / rf_inter : 128;
new_last += !new_last;
2010-05-18 17:58:33 +02:00
new_gf_alt = (rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME])
? (rfct[GOLDEN_FRAME] * 255) /
(rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME]) : 128;
new_gf_alt += !new_gf_alt;
2010-05-18 17:58:33 +02:00
// new costs
ref_frame_cost[INTRA_FRAME] = vp8_cost_zero(new_intra);
ref_frame_cost[LAST_FRAME] = vp8_cost_one(new_intra)
+ vp8_cost_zero(new_last);
ref_frame_cost[GOLDEN_FRAME] = vp8_cost_one(new_intra)
+ vp8_cost_one(new_last)
+ vp8_cost_zero(new_gf_alt);
2010-05-18 17:58:33 +02:00
ref_frame_cost[ALTREF_FRAME] = vp8_cost_one(new_intra)
+ vp8_cost_one(new_last)
+ vp8_cost_one(new_gf_alt);
2010-05-18 17:58:33 +02:00
newtotal =
rfct[INTRA_FRAME] * ref_frame_cost[INTRA_FRAME] +
rfct[LAST_FRAME] * ref_frame_cost[LAST_FRAME] +
rfct[GOLDEN_FRAME] * ref_frame_cost[GOLDEN_FRAME] +
rfct[ALTREF_FRAME] * ref_frame_cost[ALTREF_FRAME];
// old costs
ref_frame_cost[INTRA_FRAME] = vp8_cost_zero(cm->prob_intra_coded);
ref_frame_cost[LAST_FRAME] = vp8_cost_one(cm->prob_intra_coded)
+ vp8_cost_zero(cm->prob_last_coded);
ref_frame_cost[GOLDEN_FRAME] = vp8_cost_one(cm->prob_intra_coded)
+ vp8_cost_one(cm->prob_last_coded)
+ vp8_cost_zero(cm->prob_gf_coded);
ref_frame_cost[ALTREF_FRAME] = vp8_cost_one(cm->prob_intra_coded)
+ vp8_cost_one(cm->prob_last_coded)
+ vp8_cost_one(cm->prob_gf_coded);
2010-05-18 17:58:33 +02:00
oldtotal =
rfct[INTRA_FRAME] * ref_frame_cost[INTRA_FRAME] +
rfct[LAST_FRAME] * ref_frame_cost[LAST_FRAME] +
rfct[GOLDEN_FRAME] * ref_frame_cost[GOLDEN_FRAME] +
rfct[ALTREF_FRAME] * ref_frame_cost[ALTREF_FRAME];
savings += (oldtotal - newtotal) / 256;
// Update the reference frame probability numbers to reflect
// the observed counts in this frame. Doing this here insures
// that if there are multiple recode iterations the baseline
// probabilities used are updated in each iteration.
cm->prob_intra_coded = new_intra;
cm->prob_last_coded = new_last;
cm->prob_gf_coded = new_gf_alt;
}
2010-05-18 17:58:33 +02:00
savings += default_coef_context_savings(cpi);
2010-05-18 17:58:33 +02:00
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
/* do not do this if not evena allowed */
if(cpi->common.txfm_mode == ALLOW_8X8)
{
int savings8x8 = 0;
do
{
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
int j = 0;
do
{
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
int k = 0;
do
{
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
/* at every context */
/* calc probs and branch cts for this frame only */
//vp8_prob new_p [ENTROPY_NODES];
//unsigned int branch_ct [ENTROPY_NODES] [2];
int t = 0; /* token/prob index */
vp8_tree_probs_from_distribution(
MAX_ENTROPY_TOKENS, vp8_coef_encodings, vp8_coef_tree,
cpi->frame_coef_probs_8x8 [i][j][k],
cpi->frame_branch_ct_8x8 [i][j][k],
cpi->coef_counts_8x8 [i][j][k],
256, 1
);
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
do
{
const unsigned int *ct = cpi->frame_branch_ct_8x8 [i][j][k][t];
const vp8_prob newp = cpi->frame_coef_probs_8x8 [i][j][k][t];
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
const vp8_prob old = cpi->common.fc.coef_probs_8x8 [i][j][k][t];
const vp8_prob upd = vp8_coef_update_probs_8x8 [i][j][k][t];
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
const int old_b = vp8_cost_branch(ct, old);
const int new_b = vp8_cost_branch(ct, newp);
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
const int update_b = 8 +
((vp8_cost_one(upd) - vp8_cost_zero(upd)) >> 8);
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
const int s = old_b - new_b - update_b;
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
if (s > 0)
savings8x8 += s;
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
}
while (++t < MAX_ENTROPY_TOKENS - 1);
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
}
while (++k < PREV_COEF_CONTEXTS);
}
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
while (++j < COEF_BANDS);
}
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
while (++i < BLOCK_TYPES);
savings += savings8x8 >> 8;
}
2010-05-18 17:58:33 +02:00
return savings;
}
static void update_coef_probs(VP8_COMP *cpi)
{
int i = 0;
vp8_writer *const w = & cpi->bc;
int update = 0;
2010-05-18 17:58:33 +02:00
vp8_clear_system_state(); //__asm emms;
/* dry run to see if there is any udpate at all needed */
2010-05-18 17:58:33 +02:00
do
{
int j = 0;
do
{
int k = 0;
int prev_coef_savings[ENTROPY_NODES] = {0};
2010-05-18 17:58:33 +02:00
do
{
//note: use result from vp8_estimate_entropy_savings, so no need to call vp8_tree_probs_from_distribution here.
/* at every context */
/* calc probs and branch cts for this frame only */
//vp8_prob new_p [ENTROPY_NODES];
//unsigned int branch_ct [ENTROPY_NODES] [2];
2010-05-18 17:58:33 +02:00
int t = 0; /* token/prob index */
//vp8_tree_probs_from_distribution(
// MAX_ENTROPY_TOKENS, vp8_coef_encodings, vp8_coef_tree,
2010-05-18 17:58:33 +02:00
// new_p, branch_ct, (unsigned int *)cpi->coef_counts [i][j][k],
// 256, 1
// );
do
{
const vp8_prob newp = cpi->frame_coef_probs [i][j][k][t];
vp8_prob *Pold = cpi->common.fc.coef_probs [i][j][k] + t;
const vp8_prob upd = vp8_coef_update_probs [i][j][k][t];
int s = prev_coef_savings[t];
int u = 0;
s = prob_update_savings(
cpi->frame_branch_ct [i][j][k][t],
*Pold, newp, upd);
if (s > 0)
u = 1;
update += u;
2010-05-18 17:58:33 +02:00
}
while (++t < ENTROPY_NODES);
2010-05-18 17:58:33 +02:00
/* Accum token counts for generation of default statistics */
}
while (++k < PREV_COEF_CONTEXTS);
}
while (++j < COEF_BANDS);
}
while (++i < BLOCK_TYPES);
/* Is coef updated at all */
if(update==0)
{
vp8_write_bit(w, 0);
}
else
{
vp8_write_bit(w, 1);
i=0;
do
{
int j = 0;
do
{
int k = 0;
int prev_coef_savings[ENTROPY_NODES] = {0};
do
{
//note: use result from vp8_estimate_entropy_savings, so no need to call vp8_tree_probs_from_distribution here.
/* at every context */
2010-05-18 17:58:33 +02:00
/* calc probs and branch cts for this frame only */
//vp8_prob new_p [ENTROPY_NODES];
//unsigned int branch_ct [ENTROPY_NODES] [2];
int t = 0; /* token/prob index */
//vp8_tree_probs_from_distribution(
// MAX_ENTROPY_TOKENS, vp8_coef_encodings, vp8_coef_tree,
// new_p, branch_ct, (unsigned int *)cpi->coef_counts [i][j][k],
// 256, 1
// );
do
{
const vp8_prob newp = cpi->frame_coef_probs [i][j][k][t];
vp8_prob *Pold = cpi->common.fc.coef_probs [i][j][k] + t;
const vp8_prob upd = vp8_coef_update_probs [i][j][k][t];
int s = prev_coef_savings[t];
int u = 0;
s = prob_update_savings(
cpi->frame_branch_ct [i][j][k][t],
*Pold, newp, upd);
if (s > 0)
u = 1;
vp8_write(w, u, upd);
#ifdef ENTROPY_STATS
++ tree_update_hist [i][j][k][t] [u];
2010-05-18 17:58:33 +02:00
#endif
if (u)
{
/* send/use new probability */
*Pold = newp;
vp8_write_literal(w, newp, 8);
}
}
while (++t < ENTROPY_NODES);
/* Accum token counts for generation of default statistics */
#ifdef ENTROPY_STATS
t = 0;
do
{
context_counters [i][j][k][t] += cpi->coef_counts [i][j][k][t];
}
while (++t < MAX_ENTROPY_TOKENS);
#endif
}
while (++k < PREV_COEF_CONTEXTS);
2010-05-18 17:58:33 +02:00
}
while (++j < COEF_BANDS);
2010-05-18 17:58:33 +02:00
}
while (++i < BLOCK_TYPES);
2010-05-18 17:58:33 +02:00
}
2010-05-18 17:58:33 +02:00
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
/* do not do this if not evena allowed */
if(cpi->common.txfm_mode == ALLOW_8X8)
{
/* dry run to see if update is necessary */
update = 0;
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
i = 0;
do
{
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
int j = 0;
do
{
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
int k = 0;
do
{
//note: use result from vp8_estimate_entropy_savings, so no need to call vp8_tree_probs_from_distribution here.
/* at every context */
/* calc probs and branch cts for this frame only */
//vp8_prob new_p [ENTROPY_NODES];
//unsigned int branch_ct [ENTROPY_NODES] [2];
int t = 0; /* token/prob index */
//vp8_tree_probs_from_distribution(
// MAX_ENTROPY_TOKENS, vp8_coef_encodings, vp8_coef_tree,
// new_p, branch_ct, (unsigned int *)cpi->coef_counts [i][j][k],
// 256, 1
// );
do
{
const unsigned int *ct = cpi->frame_branch_ct_8x8 [i][j][k][t];
const vp8_prob newp = cpi->frame_coef_probs_8x8 [i][j][k][t];
vp8_prob *Pold = cpi->common.fc.coef_probs_8x8 [i][j][k] + t;
const vp8_prob old = *Pold;
const vp8_prob upd = vp8_coef_update_probs_8x8 [i][j][k][t];
const int old_b = vp8_cost_branch(ct, old);
const int new_b = vp8_cost_branch(ct, newp);
const int update_b = 8 +
((vp8_cost_one(upd) - vp8_cost_zero(upd)) >> 8);
const int s = old_b - new_b - update_b;
const int u = s > 0 ? 1 : 0;
#ifdef ENTROPY_STATS
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
++ tree_update_hist_8x8 [i][j][k][t] [u];
#endif
update += u;
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
}
while (++t < MAX_ENTROPY_TOKENS - 1);
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
/* Accum token counts for generation of default statistics */
#ifdef ENTROPY_STATS
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
t = 0;
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
do
{
context_counters_8x8 [i][j][k][t] += cpi->coef_counts_8x8 [i][j][k][t];
}
while (++t < MAX_ENTROPY_TOKENS);
#endif
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
}
while (++k < PREV_COEF_CONTEXTS);
}
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
while (++j < COEF_BANDS);
}
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
while (++i < BLOCK_TYPES);
if(update == 0)
{
vp8_write_bit(w, 0);
}
else
{
vp8_write_bit(w, 1);
i = 0;
do
{
int j = 0;
do
{
int k = 0;
do
{
//note: use result from vp8_estimate_entropy_savings, so no need to call vp8_tree_probs_from_distribution here.
/* at every context */
/* calc probs and branch cts for this frame only */
//vp8_prob new_p [ENTROPY_NODES];
//unsigned int branch_ct [ENTROPY_NODES] [2];
int t = 0; /* token/prob index */
//vp8_tree_probs_from_distribution(
// MAX_ENTROPY_TOKENS, vp8_coef_encodings, vp8_coef_tree,
// new_p, branch_ct, (unsigned int *)cpi->coef_counts [i][j][k],
// 256, 1
// );
do
{
const unsigned int *ct = cpi->frame_branch_ct_8x8 [i][j][k][t];
const vp8_prob newp = cpi->frame_coef_probs_8x8 [i][j][k][t];
vp8_prob *Pold = cpi->common.fc.coef_probs_8x8 [i][j][k] + t;
const vp8_prob old = *Pold;
const vp8_prob upd = vp8_coef_update_probs_8x8 [i][j][k][t];
const int old_b = vp8_cost_branch(ct, old);
const int new_b = vp8_cost_branch(ct, newp);
const int update_b = 8 +
((vp8_cost_one(upd) - vp8_cost_zero(upd)) >> 8);
const int s = old_b - new_b - update_b;
const int u = s > 0 ? 1 : 0;
vp8_write(w, u, upd);
#ifdef ENTROPY_STATS
++ tree_update_hist_8x8 [i][j][k][t] [u];
#endif
if (u)
{
/* send/use new probability */
*Pold = newp;
vp8_write_literal(w, newp, 8);
}
}
while (++t < MAX_ENTROPY_TOKENS - 1);
/* Accum token counts for generation of default statistics */
#ifdef ENTROPY_STATS
t = 0;
do
{
context_counters_8x8 [i][j][k][t] += cpi->coef_counts_8x8 [i][j][k][t];
}
while (++t < MAX_ENTROPY_TOKENS);
#endif
}
while (++k < PREV_COEF_CONTEXTS);
}
while (++j < COEF_BANDS);
}
while (++i < BLOCK_TYPES);
}
}
2010-05-18 17:58:33 +02:00
}
#ifdef PACKET_TESTING
FILE *vpxlogc = 0;
#endif
static void put_delta_q(vp8_writer *bc, int delta_q)
{
if (delta_q != 0)
{
vp8_write_bit(bc, 1);
vp8_write_literal(bc, abs(delta_q), 4);
if (delta_q < 0)
vp8_write_bit(bc, 1);
else
vp8_write_bit(bc, 0);
}
else
vp8_write_bit(bc, 0);
}
#if CONFIG_QIMODE
extern const unsigned int kf_y_mode_cts[8][VP8_YMODES];
static void decide_kf_ymode_entropy(VP8_COMP *cpi)
{
int mode_cost[MB_MODE_COUNT];
int cost;
int bestcost = INT_MAX;
int bestindex = 0;
int i, j;
for(i=0; i<8; i++)
{
vp8_cost_tokens(mode_cost, cpi->common.kf_ymode_prob[i], vp8_kf_ymode_tree);
cost = 0;
for(j=0;j<VP8_YMODES;j++)
{
cost += mode_cost[j] * cpi->ymode_count[j];
}
if(cost < bestcost)
{
bestindex = i;
bestcost = cost;
}
}
cpi->common.kf_ymode_probs_index = bestindex;
}
#endif
static segment_reference_frames(VP8_COMP *cpi)
{
VP8_COMMON *oci = &cpi->common;
MODE_INFO *mi = oci->mi;
int ref[MAX_MB_SEGMENTS]={0};
int i,j;
int mb_index=0;
MACROBLOCKD *const xd = & cpi->mb.e_mbd;
for (i = 0; i < oci->mb_rows; i++)
{
for (j = 0; j < oci->mb_cols; j++, mb_index++)
{
ref[mi[mb_index].mbmi.segment_id]|=(1<<mi[mb_index].mbmi.ref_frame);
}
mb_index++;
}
for (i = 0; i < MAX_MB_SEGMENTS; i++)
{
enable_segfeature(xd,i,SEG_LVL_REF_FRAME);
set_segdata( xd,i, SEG_LVL_REF_FRAME, ref[i]);
}
2010-05-18 17:58:33 +02:00
}
2010-05-18 17:58:33 +02:00
void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest, unsigned long *size)
{
int i, j;
VP8_HEADER oh;
VP8_COMMON *const pc = & cpi->common;
vp8_writer *const bc = & cpi->bc;
MACROBLOCKD *const xd = & cpi->mb.e_mbd;
int extra_bytes_packed = 0;
unsigned char *cx_data = dest;
oh.show_frame = (int) pc->show_frame;
oh.type = (int)pc->frame_type;
oh.version = pc->version;
oh.first_partition_length_in_bytes = 0;
2010-05-18 17:58:33 +02:00
cx_data += 3;
#if defined(SECTIONBITS_OUTPUT)
Sectionbits[active_section = 1] += sizeof(VP8_HEADER) * 8 * 256;
#endif
//vp8_kf_default_bmode_probs() is called in vp8_setup_key_frame() once for each
//K frame before encode frame. pc->kf_bmode_prob doesn't get changed anywhere
//else. No need to call it again here. --yw
//vp8_kf_default_bmode_probs( pc->kf_bmode_prob);
// every keyframe send startcode, width, height, scale factor, clamp and color type
if (oh.type == KEY_FRAME)
{
int v;
2010-05-18 17:58:33 +02:00
// Start / synch code
cx_data[0] = 0x9D;
cx_data[1] = 0x01;
cx_data[2] = 0x2a;
v = (pc->horiz_scale << 14) | pc->Width;
cx_data[3] = v;
cx_data[4] = v >> 8;
v = (pc->vert_scale << 14) | pc->Height;
cx_data[5] = v;
cx_data[6] = v >> 8;
2010-05-18 17:58:33 +02:00
extra_bytes_packed = 7;
cx_data += extra_bytes_packed ;
vp8_start_encode(bc, cx_data);
// signal clr type
vp8_write_bit(bc, pc->clr_type);
vp8_write_bit(bc, pc->clamp_type);
}
else
vp8_start_encode(bc, cx_data);
2010-05-18 17:58:33 +02:00
// Signal whether or not Segmentation is enabled
vp8_write_bit(bc, (xd->segmentation_enabled) ? 1 : 0);
// Indicate which features are enabled
if ( xd->segmentation_enabled )
2010-05-18 17:58:33 +02:00
{
// Indicate whether or not the segmentation map is being updated.
2010-05-18 17:58:33 +02:00
vp8_write_bit(bc, (xd->update_mb_segmentation_map) ? 1 : 0);
// If it is, then indicate the method that will be used.
if ( xd->update_mb_segmentation_map )
vp8_write_bit(bc, (pc->temporal_update) ? 1:0);
2010-05-18 17:58:33 +02:00
vp8_write_bit(bc, (xd->update_mb_segmentation_data) ? 1 : 0);
//segment_reference_frames(cpi);
2010-05-18 17:58:33 +02:00
if (xd->update_mb_segmentation_data)
{
signed char Data;
vp8_write_bit(bc, (xd->mb_segment_abs_delta) ? 1 : 0);
2010-05-18 17:58:33 +02:00
// For each segments id...
for (i = 0; i < MAX_MB_SEGMENTS; i++)
{
// For each segmentation codable feature...
for (j = 0; j < SEG_LVL_MAX; j++)
2010-05-18 17:58:33 +02:00
{
Data = get_segdata( xd, i, j );
2010-05-18 17:58:33 +02:00
#if CONFIG_FEATUREUPDATES
// check if there's an update
if(segfeature_changed( xd,i,j) )
{
vp8_write_bit(bc, 1);
if ( segfeature_active( xd, i, j ) )
{
// this bit is to say we are still
// active/ if we were inactive
// this is unnecessary
if ( old_segfeature_active( xd, i, j ))
{
vp8_write_bit(bc, 1);
}
// Is the segment data signed..
if ( is_segfeature_signed(j) )
{
// Encode the relevant feature data
if (Data < 0)
{
Data = - Data;
vp8_write_literal(bc, Data,
seg_feature_data_bits(j));
vp8_write_bit(bc, 1);
}
else
{
vp8_write_literal(bc, Data,
seg_feature_data_bits(j));
vp8_write_bit(bc, 0);
}
}
// Unsigned data element so no sign bit needed
else
vp8_write_literal(bc, Data,
seg_feature_data_bits(j));
}
// feature is inactive now
else if ( old_segfeature_active( xd, i, j ))
{
vp8_write_bit(bc, 0);
}
}
else
{
vp8_write_bit(bc,0);
}
#else
// If the feature is enabled...
if ( segfeature_active( xd, i, j ) )
2010-05-18 17:58:33 +02:00
{
vp8_write_bit(bc, 1);
// Is the segment data signed..
if ( is_segfeature_signed(j) )
2010-05-18 17:58:33 +02:00
{
// Encode the relevant feature data
if (Data < 0)
{
Data = - Data;
vp8_write_literal(bc, Data,
seg_feature_data_bits(j));
vp8_write_bit(bc, 1);
}
else
{
vp8_write_literal(bc, Data,
seg_feature_data_bits(j));
vp8_write_bit(bc, 0);
}
2010-05-18 17:58:33 +02:00
}
// Unsigned data element so no sign bit needed
2010-05-18 17:58:33 +02:00
else
vp8_write_literal(bc, Data,
seg_feature_data_bits(j));
2010-05-18 17:58:33 +02:00
}
else
vp8_write_bit(bc, 0);
#endif
2010-05-18 17:58:33 +02:00
}
}
}
#if CONFIG_FEATUREUPDATES
// save the segment info for updates next frame
save_segment_info ( xd );
#endif
2010-05-18 17:58:33 +02:00
if (xd->update_mb_segmentation_map)
{
Further work on Segmentation Experiment: This check in includes quite a lot of clean up and refactoring. Most of the analysis and set up for the different coding options for the segment map (currently simple distribution based coding or temporaly predicted coding), has been moved to one location (the function choose_segmap_coding_method() in segmenation.c). This code was previously scattered around in various locations making integration with other experiments and modification / debug more difficult. Currently the functionality is as it was with the exception that the prediction probabilities are now only transmitted when the temporal prediction mode is selected. There is still quite a bit more clean up work that will be possible when the #ifdef is removed. Also at that time I may rename and alter the sense of macroblock based variable "segment_flag" which indicates (1 that the segmnet id is not predicted vs 0 that it is predicted). I also intend to experiment with a spatial prediction mode that can be used when coding a key frame segment map or in cases where temporal prediction does not work well but there is spatial correlation. In a later check in when the ifdefs have gone I may also move the call to choose_segmap_coding_method() to just before where the bitsream is packed (currently it is in vp8_encode_frame()) to further reduce the possibility of clashes with other experiments and prevent it being called on each itteration of the recode loop. Change-Id: I3d4aba2a2826ec21f367678d5b07c1d1c36db168
2011-11-15 12:13:33 +01:00
// Send the tree probabilities used to decode unpredicted
// macro-block segments
2010-05-18 17:58:33 +02:00
for (i = 0; i < MB_FEATURE_TREE_PROBS; i++)
{
int Data = xd->mb_segment_tree_probs[i];
if (Data != 255)
{
vp8_write_bit(bc, 1);
vp8_write_literal(bc, Data, 8);
}
else
vp8_write_bit(bc, 0);
}
Further work on Segmentation Experiment: This check in includes quite a lot of clean up and refactoring. Most of the analysis and set up for the different coding options for the segment map (currently simple distribution based coding or temporaly predicted coding), has been moved to one location (the function choose_segmap_coding_method() in segmenation.c). This code was previously scattered around in various locations making integration with other experiments and modification / debug more difficult. Currently the functionality is as it was with the exception that the prediction probabilities are now only transmitted when the temporal prediction mode is selected. There is still quite a bit more clean up work that will be possible when the #ifdef is removed. Also at that time I may rename and alter the sense of macroblock based variable "segment_flag" which indicates (1 that the segmnet id is not predicted vs 0 that it is predicted). I also intend to experiment with a spatial prediction mode that can be used when coding a key frame segment map or in cases where temporal prediction does not work well but there is spatial correlation. In a later check in when the ifdefs have gone I may also move the call to choose_segmap_coding_method() to just before where the bitsream is packed (currently it is in vp8_encode_frame()) to further reduce the possibility of clashes with other experiments and prevent it being called on each itteration of the recode loop. Change-Id: I3d4aba2a2826ec21f367678d5b07c1d1c36db168
2011-11-15 12:13:33 +01:00
// If predictive coding of segment map is enabled send the
// prediction probabilities.
if ( pc->temporal_update )
Further work on Segmentation Experiment: This check in includes quite a lot of clean up and refactoring. Most of the analysis and set up for the different coding options for the segment map (currently simple distribution based coding or temporaly predicted coding), has been moved to one location (the function choose_segmap_coding_method() in segmenation.c). This code was previously scattered around in various locations making integration with other experiments and modification / debug more difficult. Currently the functionality is as it was with the exception that the prediction probabilities are now only transmitted when the temporal prediction mode is selected. There is still quite a bit more clean up work that will be possible when the #ifdef is removed. Also at that time I may rename and alter the sense of macroblock based variable "segment_flag" which indicates (1 that the segmnet id is not predicted vs 0 that it is predicted). I also intend to experiment with a spatial prediction mode that can be used when coding a key frame segment map or in cases where temporal prediction does not work well but there is spatial correlation. In a later check in when the ifdefs have gone I may also move the call to choose_segmap_coding_method() to just before where the bitsream is packed (currently it is in vp8_encode_frame()) to further reduce the possibility of clashes with other experiments and prevent it being called on each itteration of the recode loop. Change-Id: I3d4aba2a2826ec21f367678d5b07c1d1c36db168
2011-11-15 12:13:33 +01:00
{
for (i = 0; i < PREDICTION_PROBS; i++)
Further work on Segmentation Experiment: This check in includes quite a lot of clean up and refactoring. Most of the analysis and set up for the different coding options for the segment map (currently simple distribution based coding or temporaly predicted coding), has been moved to one location (the function choose_segmap_coding_method() in segmenation.c). This code was previously scattered around in various locations making integration with other experiments and modification / debug more difficult. Currently the functionality is as it was with the exception that the prediction probabilities are now only transmitted when the temporal prediction mode is selected. There is still quite a bit more clean up work that will be possible when the #ifdef is removed. Also at that time I may rename and alter the sense of macroblock based variable "segment_flag" which indicates (1 that the segmnet id is not predicted vs 0 that it is predicted). I also intend to experiment with a spatial prediction mode that can be used when coding a key frame segment map or in cases where temporal prediction does not work well but there is spatial correlation. In a later check in when the ifdefs have gone I may also move the call to choose_segmap_coding_method() to just before where the bitsream is packed (currently it is in vp8_encode_frame()) to further reduce the possibility of clashes with other experiments and prevent it being called on each itteration of the recode loop. Change-Id: I3d4aba2a2826ec21f367678d5b07c1d1c36db168
2011-11-15 12:13:33 +01:00
{
int Data = pc->segment_pred_probs[i];
Further work on Segmentation Experiment: This check in includes quite a lot of clean up and refactoring. Most of the analysis and set up for the different coding options for the segment map (currently simple distribution based coding or temporaly predicted coding), has been moved to one location (the function choose_segmap_coding_method() in segmenation.c). This code was previously scattered around in various locations making integration with other experiments and modification / debug more difficult. Currently the functionality is as it was with the exception that the prediction probabilities are now only transmitted when the temporal prediction mode is selected. There is still quite a bit more clean up work that will be possible when the #ifdef is removed. Also at that time I may rename and alter the sense of macroblock based variable "segment_flag" which indicates (1 that the segmnet id is not predicted vs 0 that it is predicted). I also intend to experiment with a spatial prediction mode that can be used when coding a key frame segment map or in cases where temporal prediction does not work well but there is spatial correlation. In a later check in when the ifdefs have gone I may also move the call to choose_segmap_coding_method() to just before where the bitsream is packed (currently it is in vp8_encode_frame()) to further reduce the possibility of clashes with other experiments and prevent it being called on each itteration of the recode loop. Change-Id: I3d4aba2a2826ec21f367678d5b07c1d1c36db168
2011-11-15 12:13:33 +01:00
if (Data != 255)
{
vp8_write_bit(bc, 1);
vp8_write_literal(bc, Data, 8);
}
else
vp8_write_bit(bc, 0);
}
}
2010-05-18 17:58:33 +02:00
}
}
// Encode the common prediction model status flag probability updates for
// the reference frame
if ( pc->frame_type != KEY_FRAME )
{
for (i = 0; i < PREDICTION_PROBS; i++)
{
if ( cpi->ref_pred_probs_update[i] )
{
vp8_write_bit(bc, 1);
vp8_write_literal(bc, pc->ref_pred_probs[i], 8);
}
else
vp8_write_bit(bc, 0);
}
}
Improved coding using 8x8 transform In summary, this commit encompasses a series of changes in attempt to improve the 8x8 transform based coding to help overall compression quality, please refer to the detailed commit history below for what are the rationale underly the series of changes: a. A frame level flag to indicate if 8x8 transform is used at all. b. 8x8 transform is not used for key frames and small image size. c. On inter coded frame, macroblocks using modes B_PRED, SPLIT_MV and I8X8_PRED are forced to using 4x4 transform based coding, the rest uses 8x8 transform based coding. d. Encoder and decoder has the same assumption on the relationship between prediction modes and transform size, therefore no signaling is encoded in bitstream. e. Mode decision process now calculate the rate and distortion scores using their respective transforms. Overall test results: 1. HD set http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120206.html (avg psnr: 3.09% glb psnr: 3.22%, ssim: 3.90%) 2. Cif set: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120206.html (avg psnr: -0.03%, glb psnr: -0.02%, ssim: -0.04%) It should be noted here, as 8x8 transform coding itself is disabled for cif size clips, the 0.03% loss is purely from the 1 bit/frame flag overhead on if 8x8 transform is used or not for the frame. ---patch history for future reference--- Patch 1: this commit tries to select transform size based on macroblock prediction mode. If the size of a prediction mode is 16x16, then the macroblock is forced to use 8x8 transform. If the prediction mode is B_PRED, SPLITMV or I8X8_PRED, then the macroblock is forced to use 4x4 transform. Tests on the following HD clips showed mixed results: (all hd clips only used first 100 frames in the test) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_log.html while the results are mixed and overall negative, it is interesting to see 8x8 helped a few of the clips. Patch 2: this patch tries to hard-wire selection of transform size based on prediction modes without using segmentation to signal the transform size. encoder and decoder both takes the same assumption that all macroblocks use 8x8 transform except when prediciton mode is B_PRED, I8X8_PRED or SPLITMV. Test results are as follows: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cifmodebase8x8_0125.html http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdmodebased8x8_0125log.html Interestingly, by removing the overhead or coding the segmentation, the results on this limited HD set have turn positive on average. Patch 3: this patch disabled the usage of 8x8 transform on key frames, and kept the logic from patch 2 for inter frames only. test results on HD set turned decidedly positive with 8x8 transform enabled on inter frame with 16x16 prediction modes: (avg psnr: .81% glb psnr: .82 ssim: .55%) http://www.corp.google.com/~yaowu/no_crawl/t8x8/hdintermode8x8_0125.html results on cif set still negative overall Patch 4: continued from last patch, but now in mode decision process, the rate and distortion estimates are computed based on 8x8 transform results for MBs with modes associated with 8x8 transform. This patch also fixed a problem related to segment based eob coding when 8x8 transform is used. The patch significantly improved the results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd8x8RDintermode.html (avg psnr: 2.70% glb psnr: 2.76% ssim: 3.34%) results on cif also improved, though they are still negative compared to baseline that uses 4x4 transform only: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif8x8RDintermode.html (avg psnr: -.78% glb psnr: -.86% ssim: -.19%) Patch 5: This patch does 3 things: a. a bunch of decoder bug fixes, encodings and decodings were verified to have matched recon buffer on a number of encodes on cif size mobile and hd version of _pedestrian. b. the patch further improved the rate distortion calculation of MBS that use 8x8 transform. This provided some further gain on compression. c. the patch also got the experimental work SEG_LVL_EOB to work with 8x8 transformed macroblock, test results indicates it improves the cif set but hurt the HD set slightly. Tests results on HD clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_20120201.html (avg psnr: 3.19% glb psnr: 3.30% ssim: 3.93%) Test results on cif clips: http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_20120201.html (avg psnr: -.47% glb psnr: -.51% ssim: +.28%) Patch 6: Added a frame level flag to indicate if 8x8 transform is allowed at all. temporarily the decision is based on frame size, can be optimized later one. This get the cif results to basically unchanged, with one bit per frame overhead on both cif and hd clips. Patch 8: Rebase and Merge to head by PGW. Fixed some suspect 4s that look like hey should be 64s in regard to segmented EOB. Perhaps #defines would be bette. Bulit and tested without T8x8 enabled and produces unchanged output. Patch 9: Corrected misalligned code/decode of "txfm_mode" bit. Limited testing for correct encode and decode with T8x8 configured on derf clips. Change-Id: I156e1405d25f81579d579dff8ab9af53944ec49c
2012-02-10 01:12:23 +01:00
vp8_write_bit(bc, pc->txfm_mode);
// Encode the loop filter level and type
2010-05-18 17:58:33 +02:00
vp8_write_bit(bc, pc->filter_type);
vp8_write_literal(bc, pc->filter_level, 6);
vp8_write_literal(bc, pc->sharpness_level, 3);
// Write out loop filter deltas applied at the MB level based on mode or ref frame (if they are enabled).
vp8_write_bit(bc, (xd->mode_ref_lf_delta_enabled) ? 1 : 0);
if (xd->mode_ref_lf_delta_enabled)
{
// Do the deltas need to be updated
int send_update = xd->mode_ref_lf_delta_update;
2010-05-18 17:58:33 +02:00
vp8_write_bit(bc, send_update);
if (send_update)
2010-05-18 17:58:33 +02:00
{
int Data;
// Send update
for (i = 0; i < MAX_REF_LF_DELTAS; i++)
{
Data = xd->ref_lf_deltas[i];
// Frame level data
if (xd->ref_lf_deltas[i] != xd->last_ref_lf_deltas[i])
2010-05-18 17:58:33 +02:00
{
xd->last_ref_lf_deltas[i] = xd->ref_lf_deltas[i];
2010-05-18 17:58:33 +02:00
vp8_write_bit(bc, 1);
if (Data > 0)
{
vp8_write_literal(bc, (Data & 0x3F), 6);
vp8_write_bit(bc, 0); // sign
}
else
{
Data = -Data;
vp8_write_literal(bc, (Data & 0x3F), 6);
vp8_write_bit(bc, 1); // sign
}
}
else
vp8_write_bit(bc, 0);
}
// Send update
for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
{
Data = xd->mode_lf_deltas[i];
if (xd->mode_lf_deltas[i] != xd->last_mode_lf_deltas[i])
2010-05-18 17:58:33 +02:00
{
xd->last_mode_lf_deltas[i] = xd->mode_lf_deltas[i];
2010-05-18 17:58:33 +02:00
vp8_write_bit(bc, 1);
if (Data > 0)
{
vp8_write_literal(bc, (Data & 0x3F), 6);
vp8_write_bit(bc, 0); // sign
}
else
{
Data = -Data;
vp8_write_literal(bc, (Data & 0x3F), 6);
vp8_write_bit(bc, 1); // sign
}
}
else
vp8_write_bit(bc, 0);
}
}
}
//signal here is multi token partition is enabled
//vp8_write_literal(bc, pc->multi_token_partition, 2);
vp8_write_literal(bc, 0, 2);
2010-05-18 17:58:33 +02:00
// Frame Q baseline quantizer index
vp8_write_literal(bc, pc->base_qindex, QINDEX_BITS);
2010-05-18 17:58:33 +02:00
// Transmit Dc, Second order and Uv quantizer delta information
put_delta_q(bc, pc->y1dc_delta_q);
put_delta_q(bc, pc->y2dc_delta_q);
put_delta_q(bc, pc->y2ac_delta_q);
put_delta_q(bc, pc->uvdc_delta_q);
put_delta_q(bc, pc->uvac_delta_q);
// When there is a key frame all reference buffers are updated using the new key frame
if (pc->frame_type != KEY_FRAME)
{
// Should the GF or ARF be updated using the transmitted frame or buffer
vp8_write_bit(bc, pc->refresh_golden_frame);
vp8_write_bit(bc, pc->refresh_alt_ref_frame);
// If not being updated from current frame should either GF or ARF be updated from another buffer
if (!pc->refresh_golden_frame)
vp8_write_literal(bc, pc->copy_buffer_to_gf, 2);
if (!pc->refresh_alt_ref_frame)
vp8_write_literal(bc, pc->copy_buffer_to_arf, 2);
// Indicate reference frame sign bias for Golden and ARF frames (always 0 for last frame buffer)
vp8_write_bit(bc, pc->ref_frame_sign_bias[GOLDEN_FRAME]);
vp8_write_bit(bc, pc->ref_frame_sign_bias[ALTREF_FRAME]);
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#if CONFIG_HIGH_PRECISION_MV
// Signal whether to allow high MV precision
vp8_write_bit(bc, (xd->allow_high_precision_mv) ? 1 : 0);
#endif
2010-05-18 17:58:33 +02:00
}
vp8_write_bit(bc, pc->refresh_entropy_probs);
if (pc->frame_type != KEY_FRAME)
vp8_write_bit(bc, pc->refresh_last_frame);
#ifdef ENTROPY_STATS
if (pc->frame_type == INTER_FRAME)
active_section = 0;
else
active_section = 7;
#endif
vp8_clear_system_state(); //__asm emms;
update_coef_probs(cpi);
#ifdef ENTROPY_STATS
active_section = 2;
#endif
// Write out the mb_no_coeff_skip flag
vp8_write_bit(bc, pc->mb_no_coeff_skip);
if (pc->frame_type == KEY_FRAME)
{
#if CONFIG_QIMODE
decide_kf_ymode_entropy(cpi);
#endif
2010-05-18 17:58:33 +02:00
write_kfmodes(cpi);
#ifdef ENTROPY_STATS
active_section = 8;
#endif
}
else
{
pack_inter_mode_mvs(cpi);
vp8_update_mode_context(&cpi->common);
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
active_section = 1;
#endif
}
2010-05-18 17:58:33 +02:00
vp8_stop_encode(bc);
oh.first_partition_length_in_bytes = cpi->bc.pos;
/* update frame tag */
{
int v = (oh.first_partition_length_in_bytes << 5) |
(oh.show_frame << 4) |
(oh.version << 1) |
oh.type;
dest[0] = v;
dest[1] = v >> 8;
dest[2] = v >> 16;
}
*size = VP8_HEADER_SIZE + extra_bytes_packed + cpi->bc.pos;
2010-05-18 17:58:33 +02:00
vp8_start_encode(&cpi->bc2, cx_data + bc->pos);
2010-05-18 17:58:33 +02:00
pack_tokens(&cpi->bc2, cpi->tok, cpi->tok_count);
2010-05-18 17:58:33 +02:00
vp8_stop_encode(&cpi->bc2);
2010-05-18 17:58:33 +02:00
*size += cpi->bc2.pos;
2010-05-18 17:58:33 +02:00
}
#ifdef ENTROPY_STATS
void print_tree_update_probs()
{
int i, j, k, l;
FILE *f = fopen("context.c", "a");
int Sum;
fprintf(f, "\n/* Update probabilities for token entropy tree. */\n\n");
fprintf(f, "const vp8_prob tree_update_probs[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES] = {\n");
2010-05-18 17:58:33 +02:00
for (i = 0; i < BLOCK_TYPES; i++)
{
fprintf(f, " { \n");
for (j = 0; j < COEF_BANDS; j++)
{
fprintf(f, " {\n");
for (k = 0; k < PREV_COEF_CONTEXTS; k++)
{
fprintf(f, " {");
for (l = 0; l < ENTROPY_NODES; l++)
2010-05-18 17:58:33 +02:00
{
Sum = tree_update_hist[i][j][k][l][0] + tree_update_hist[i][j][k][l][1];
if (Sum > 0)
{
if (((tree_update_hist[i][j][k][l][0] * 255) / Sum) > 0)
fprintf(f, "%3ld, ", (tree_update_hist[i][j][k][l][0] * 255) / Sum);
else
fprintf(f, "%3ld, ", 1);
}
else
fprintf(f, "%3ld, ", 128);
}
fprintf(f, "},\n");
}
fprintf(f, " },\n");
}
fprintf(f, " },\n");
}
fprintf(f, "};\n");
fprintf(f, "const vp8_prob tree_update_probs_8x8[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES] = {\n");
for (i = 0; i < BLOCK_TYPES; i++)
{
fprintf(f, " { \n");
for (j = 0; j < COEF_BANDS; j++)
{
fprintf(f, " {\n");
for (k = 0; k < PREV_COEF_CONTEXTS; k++)
{
fprintf(f, " {");
for (l = 0; l < MAX_ENTROPY_TOKENS - 1; l++)
{
Sum = tree_update_hist_8x8[i][j][k][l][0] + tree_update_hist_8x8[i][j][k][l][1];
if (Sum > 0)
{
if (((tree_update_hist_8x8[i][j][k][l][0] * 255) / Sum) > 0)
fprintf(f, "%3ld, ", (tree_update_hist_8x8[i][j][k][l][0] * 255) / Sum);
else
fprintf(f, "%3ld, ", 1);
}
else
fprintf(f, "%3ld, ", 128);
}
fprintf(f, "},\n");
}
fprintf(f, " },\n");
}
fprintf(f, " },\n");
}
2010-05-18 17:58:33 +02:00
fclose(f);
}
#endif