potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 15:53:03 +04:00
Code Packages Releases Wiki Activity

softcascade: remove device dependency on gpu

Browse Source
This commit is contained in:
marina.kolpakova
2013-03-03 16:27:49 +04:00
parent 6daf17f974
commit 3c8e66d580
9 changed files with 77 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/softcascade/src/cuda/icf-sc.cu
+27 -12
View File
@@ -40,13 +40,28 @@
//
//M*/
#include <opencv2/gpu/device/common.hpp>
#include <opencv2/gpu/device/saturate_cast.hpp>
#include <cuda_invoker.hpp>
#include <float.h>
#include <stdio.h>
namespace
{
#if defined(__GNUC__)
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__, __func__)
#else /* defined(__CUDACC__) || defined(__MSVC__) */
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__)
#endif
inline void ___cudaSafeCall(cudaError_t err, const char *file, const int line, const char *func = "")
{
//if (cudaSuccess != err) cv::gpu::error(cudaGetErrorString(err), file, line, func);
}
}
#ifndef CV_PI
#define CV_PI 3.1415926535897932384626433832795
#endif
namespace cv { namespace softcascade { namespace device {
typedef unsigned char uchar;
@@ -126,7 +141,7 @@ typedef unsigned char uchar;
luvg[luvgPitch * (y + 2 * 480) + x] = v;
}
void bgr2Luv(const PtrStepSzb& bgr, PtrStepSzb luv)
void bgr2Luv(const cv::gpu::PtrStepSzb& bgr, cv::gpu::PtrStepSzb luv)
{
dim3 block(32, 8);
dim3 grid(bgr.cols / 32, bgr.rows / 8);
@@ -208,7 +223,7 @@ typedef unsigned char uchar;
texture<uchar, cudaTextureType2D, cudaReadModeElementType> tgray;
template<bool isDefaultNum>
__global__ void gray2hog(PtrStepSzb mag)
__global__ void gray2hog(cv::gpu::PtrStepSzb mag)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -223,7 +238,7 @@ typedef unsigned char uchar;
mag( 480 * fast_angle_bin<isDefaultNum>(dy, dx) + y, x) = cmag;
}
void gray2hog(const PtrStepSzb& gray, PtrStepSzb mag, const int bins)
void gray2hog(const cv::gpu::PtrStepSzb& gray, cv::gpu::PtrStepSzb mag, const int bins)
{
dim3 block(32, 8);
dim3 grid(gray.cols / 32, gray.rows / 8);
@@ -326,8 +341,8 @@ typedef unsigned char uchar;
}
}
void suppress(const PtrStepSzb& objects, PtrStepSzb overlaps, PtrStepSzi ndetections,
PtrStepSzb suppressed, cudaStream_t stream)
void suppress(const cv::gpu::PtrStepSzb& objects, cv::gpu::PtrStepSzb overlaps, cv::gpu::PtrStepSzi ndetections,
cv::gpu::PtrStepSzb suppressed, cudaStream_t stream)
{
int block = 192;
int grid = 1;
@@ -529,8 +544,8 @@ __global__ void soft_cascade(const CascadeInvoker<Policy> invoker, Detection* ob
}
template<typename Policy>
void CascadeInvoker<Policy>::operator()(const PtrStepSzb& roi, const PtrStepSzi& hogluv,
PtrStepSz<uchar4> objects, const int downscales, const cudaStream_t& stream) const
void CascadeInvoker<Policy>::operator()(const cv::gpu::PtrStepSzb& roi, const cv::gpu::PtrStepSzi& hogluv,
cv::gpu::PtrStepSz<uchar4> objects, const int downscales, const cudaStream_t& stream) const
{
int fw = roi.rows;
int fh = roi.cols;
@@ -562,7 +577,7 @@ void CascadeInvoker<Policy>::operator()(const PtrStepSzb& roi, const PtrStepSzi&
}
}
template void CascadeInvoker<GK107PolicyX4>::operator()(const PtrStepSzb& roi, const PtrStepSzi& hogluv,
PtrStepSz<uchar4> objects, const int downscales, const cudaStream_t& stream) const;
template void CascadeInvoker<GK107PolicyX4>::operator()(const cv::gpu::PtrStepSzb& roi, const cv::gpu::PtrStepSzi& hogluv,
cv::gpu::PtrStepSz<uchar4> objects, const int downscales, const cudaStream_t& stream) const;
}}}
modules/softcascade/src/cuda_invoker.hpp
+5 -3
View File
@@ -44,9 +44,9 @@
#ifndef __OPENCV_ICF_HPP__
#define __OPENCV_ICF_HPP__
#include <opencv2/gpu/device/common.hpp>
using namespace cv::gpu::device;
// #include <opencv2/gpu/device/common.hpp>
#include "opencv2/core/cuda_devptrs.hpp"
#include "cuda_runtime_api.h"
#if defined __CUDACC__
# define __device_inline__ __device__ __forceinline__
@@ -57,6 +57,8 @@ using namespace cv::gpu::device;
namespace cv { namespace softcascade { namespace device {
typedef unsigned char uchar;
struct Octave
{
ushort index;
modules/softcascade/src/detector_cuda.cpp
+19 -5
View File
@@ -41,9 +41,9 @@
//M*/
#include "precomp.hpp"
#include "opencv2/gpu/stream_accessor.hpp"
#if !defined (HAVE_CUDA)
#define throw_nogpu() CV_Error(CV_GpuNotSupported, "The library is compiled without CUDA support")
cv::softcascade::SCascade::SCascade(const double, const double, const int, const int) { throw_nogpu(); }
cv::softcascade::SCascade::~SCascade() { throw_nogpu(); }
@@ -54,14 +54,28 @@ void cv::softcascade::SCascade::detect(InputArray, InputArray, OutputArray, cv::
void cv::softcascade::SCascade::read(const FileNode& fn) { Algorithm::read(fn); }
cv::gpu::ChannelsProcessor::ChannelsProcessor() { throw_nogpu(); }
cv::gpu::ChannelsProcessor::~ChannelsProcessor() { throw_nogpu(); }
cv::softcascade::ChannelsProcessor::ChannelsProcessor() { throw_nogpu(); }
cv::softcascade::ChannelsProcessor::~ChannelsProcessor() { throw_nogpu(); }
cv::Ptr<cv::gpu::ChannelsProcessor> cv::gpu::ChannelsProcessor::create(const int, const int, const int)
{ throw_nogpu(); return cv::Ptr<cv::gpu::ChannelsProcessor>(0); }
cv::Ptr<cv::softcascade::ChannelsProcessor> cv::softcascade::ChannelsProcessor::create(const int, const int, const int)
{ throw_nogpu(); return cv::Ptr<cv::softcascade::ChannelsProcessor>(0); }
#else
# include "cuda_invoker.hpp"
# include "opencv2/core/stream_accessor.hpp"
namespace
{
#if defined(__GNUC__)
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__, __func__)
#else /* defined(__CUDACC__) || defined(__MSVC__) */
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__)
#endif
inline void ___cudaSafeCall(cudaError_t err, const char *file, const int line, const char *func = "")
{
//if (cudaSuccess != err) cv::gpu::error(cudaGetErrorString(err), file, line, func);
}
}
cv::softcascade::device::Level::Level(int idx, const Octave& oct, const float scale, const int w, const int h)
: octave(idx), step(oct.stages), relScale(scale / oct.scale)