generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added OpenCL optimization for LBP-based face detector

Browse Source
This commit is contained in:
Vadim Pisarevsky
2013-12-23 15:28:50 +04:00
parent 1540910542
commit d084d19779
3 changed files with 102 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
modules/objdetect/src/opencl/cascadedetect.cl
View File

@@ -105,7 +105,7 @@ __kernel void runHaarClassifierStump(
}
}
#if 0
__kernel void runLBPClassifierStump(
__global const int* sum,
int sumstep, int sumoffset,
@@ -119,45 +119,48 @@ __kernel void runLBPClassifierStump(
volatile __global int* facepos,
int2 imgsize, int xyscale, float factor,
int4 normrect, int2 windowsize, int maxFaces)
int2 windowsize, int maxFaces)
{
int ix = get_global_id(0)*xyscale;
int iy = get_global_id(1)*xyscale;
sumstep /= sizeof(int);
sqsumstep /= sizeof(int);
if( ix < imgsize.x && iy < imgsize.y )
{
int stageIdx;
__global const Stump* stump = stumps;
__global const int* p = sum + mad24(iy, sumstep, ix);
for( stageIdx = 0; stageIdx < nstages; stageIdx++ )
{
int i, ntrees = stages[stageIdx].ntrees;
float s = 0.f;
for( i = 0; i < ntrees; i++, stump++ )
for( i = 0; i < ntrees; i++, stump++, bitsets += bitsetSize )
{
float4 st = stump->st;
__global const OptLBPFeature* f = optfeatures + as_int(st.x);
int16 ofs = f->ofs;
#define CALC_SUM_OFS_(p0, p1, p2, p3, ptr) \
((ptr)[p0] - (ptr)[p1] - (ptr)[p2] + (ptr)[p3])
int cval = CALC_SUM_OFS_( ofs.s5, ofs.s6, ofs.s9, ofs.sa, p );
int4 ofs = f->ofs[0];
sval = (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.x;
ofs = f->ofs[1];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.y;
if( weight.z > 0 )
{
ofs = f->ofs[2];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.z;
}
int mask, idx = (CALC_SUM_OFS_( ofs.s0, ofs.s1, ofs.s4, ofs.s5, p ) >= cval ? 4 : 0); // 0
idx |= (CALC_SUM_OFS_( ofs.s1, ofs.s2, ofs.s5, ofs.s6, p ) >= cval ? 2 : 0); // 1
idx |= (CALC_SUM_OFS_( ofs.s2, ofs.s3, ofs.s6, ofs.s7, p ) >= cval ? 1 : 0); // 2
s += (sval < st.y*nf) ? st.z : st.w;
mask = (CALC_SUM_OFS_( ofs.s6, ofs.s7, ofs.sa, ofs.sb, p ) >= cval ? 16 : 0); // 5
mask |= (CALC_SUM_OFS_( ofs.sa, ofs.sb, ofs.se, ofs.sf, p ) >= cval ? 8 : 0); // 8
mask |= (CALC_SUM_OFS_( ofs.s9, ofs.sa, ofs.sd, ofs.se, p ) >= cval ? 4 : 0); // 7
mask |= (CALC_SUM_OFS_( ofs.s8, ofs.s9, ofs.sc, ofs.sd, p ) >= cval ? 2 : 0); // 6
mask |= (CALC_SUM_OFS_( ofs.s4, ofs.s5, ofs.s8, ofs.s9, p ) >= cval ? 1 : 0); // 7
s += (bitsets[idx] & (1 << mask)) ? st.z : st.w;
}
if( s < stages[stageIdx].threshold )
break;
break;
}
if( stageIdx == nstages )
@@ -174,4 +177,4 @@ __kernel void runLBPClassifierStump(
}
}
}
#endif