int st = level.octave * level.step;
const int stEnd = st + level.step;
+ const int hogluvStep = gridDim.y * Policy::STA_Y;
float confidence = 0.f;
for(; st < stEnd; st += Policy::WARP)
{
Node node = nodes[nId];
float threshold = rescale<isUp>(level, node);
- int sum = get<isUp>(x, y + (node.threshold >> 28) * 120, node.rect);
+ int sum = get<isUp>(x, y + (node.threshold >> 28) * hogluvStep, node.rect);
int next = 1 + (int)(sum >= threshold);
node = nodes[nId + next];
threshold = rescale<isUp>(level, node);
- sum = get<isUp>(x, y + (node.threshold >> 28) * 120, node.rect);
+ sum = get<isUp>(x, y + (node.threshold >> 28) * hogluvStep, node.rect);
const int lShift = (next - 1) * 2 + (int)(sum >= threshold);
float impact = leaves[(st + threadIdx.x) * 4 + lShift];
{
CascadeInvoker(): levels(0), stages(0), nodes(0), leaves(0), scales(0) {}
- CascadeInvoker(const PtrStepSzb& _levels, const PtrStepSzb& _octaves, const PtrStepSzf& _stages,
+ CascadeInvoker(const PtrStepSzb& _levels, const PtrStepSzf& _stages,
const PtrStepSzb& _nodes, const PtrStepSzf& _leaves)
: levels((const Level*)_levels.ptr()),
stages((const float*)_stages.ptr()),
cudaMemset(count.data, 0, sizeof(Detection));
cudaSafeCall( cudaGetLastError());
device::icf::CascadeInvoker<device::icf::GK107PolicyX4> invoker
- = device::icf::CascadeInvoker<device::icf::GK107PolicyX4>(levels, octaves, stages, nodes, leaves);
+ = device::icf::CascadeInvoker<device::icf::GK107PolicyX4>(levels, stages, nodes, leaves);
invoker(roi, hogluv, objects, count, downscales, stream);
}
void integrate(const int fh, const int fw, Stream& s)
{
GpuMat channels(plane, cv::Rect(0, 0, fw, fh * Fields::HOG_LUV_BINS));
- cv::gpu::resize(channels, shrunk, cv::Size(), 0.25, 0.25, CV_INTER_AREA, s);
+ cv::gpu::resize(channels, shrunk, cv::Size(), 1.f / shrinkage, 1.f / shrinkage, CV_INTER_AREA, s);
if (info.majorVersion() < 3)
cv::gpu::integralBuffered(shrunk, hogluv, integralBuffer, s);
GpuMat tmp = GpuMat(objects, cv::Rect(0, 0, sizeof(Detection), 1));
- objects = GpuMat(objects, cv::Rect( sizeof(Detection), 0, objects.cols - sizeof(Detection), 1));
+ objects = GpuMat(objects, cv::Rect( sizeof(Detection), 0, objects.cols - sizeof(Detection), 1));
cudaStream_t stream = StreamAccessor::getStream(s);
flds.detect(rois, tmp, objects, stream);
void cv::gpu::SCascade::genRoi(InputArray _roi, OutputArray _mask, Stream& stream) const
{
CV_Assert(fields);
+ int shr = (*fields).shrinkage;
const GpuMat roi = _roi.getGpuMat();
- _mask.create( roi.cols / 4, roi.rows / 4, roi.type() );
+ _mask.create( roi.cols / shr, roi.rows / shr, roi.type() );
GpuMat mask = _mask.getGpuMat();
cv::gpu::GpuMat tmp;
- cv::gpu::resize(roi, tmp, cv::Size(), 0.25, 0.25, CV_INTER_AREA, stream);
+ cv::gpu::resize(roi, tmp, cv::Size(), 1.f / shr, 1.f / shr, CV_INTER_AREA, stream);
cv::gpu::transpose(tmp, mask, stream);
}