Fix rate estimate issue in transform block partition coding
This commit fixes the over count issue in the recursive transform block partition rate cost estimation. It improves the compression performance by about 0.45%. Change-Id: I01ccda954ed0e120263977472c1c759c3c67170c
This commit is contained in:
@@ -1233,7 +1233,8 @@ static void select_tx_block(const VP9_COMP *cpi, MACROBLOCK *x,
|
||||
if (cpi->common.tx_mode == TX_MODE_SELECT || tx_size == TX_4X4) {
|
||||
tx_block_rd_b(x, tx_size, blk_row, blk_col, plane, block,
|
||||
plane_bsize, ta, tl, rate, dist, bsse, skip);
|
||||
*rate += 256;
|
||||
if (tx_size > TX_4X4)
|
||||
*rate += 256;
|
||||
this_rd = RDCOST(x->rdmult, x->rddiv, *rate, *dist);
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user