{\r
namespace lbp\r
{\r
- int classifyStump(const DevMem2Db mstages,\r
- const int nstages,\r
- const DevMem2Di mnodes,\r
- const DevMem2Df mleaves,\r
- const DevMem2Di msubsets,\r
- const DevMem2Db mfeatures,\r
- const DevMem2Di integral,\r
- const int workWidth,\r
- const int workHeight,\r
- const int clWidth,\r
- const int clHeight,\r
- float scale,\r
- int step,\r
- int subsetSize,\r
- DevMem2D_<int4> objects);\r
+ classifyStump(const DevMem2Db mstages,\r
+ const int nstages,\r
+ const DevMem2Di mnodes,\r
+ const DevMem2Df mleaves,\r
+ const DevMem2Di msubsets,\r
+ const DevMem2Db mfeatures,\r
+ const DevMem2Di integral,\r
+ const int workWidth,\r
+ const int workHeight,\r
+ const int clWidth,\r
+ const int clHeight,\r
+ float scale,\r
+ int step,\r
+ int subsetSize,\r
+ DevMem2D_<int4> objects,\r
+ unsigned int* classified);\r
}\r
}}}\r
\r
maxObjectSize = image.size();\r
\r
scaledImageBuffer.create(image.rows + 1, image.cols + 1, CV_8U);\r
+ unsigned int* classified = new unsigned int[1];\r
+ *classified = 0;\r
+ unsigned int* dclassified;\r
+ cudaMalloc(&dclassified, sizeof(int));\r
+ cudaMemcpy(dclassified, classified, sizeof(int), cudaMemcpyHostToDevice);\r
\r
for( double factor = 1; ; factor *= scaleFactor )\r
{\r
\r
int step = (factor <= 2.) + 1;\r
\r
- int res = cv::gpu::device::lbp::classifyStump(stage_mat, stage_mat.cols / sizeof(Stage), nodes_mat, leaves_mat, subsets_mat, features_mat,\r
- integral, processingRectSize.width, processingRectSize.height, windowSize.width, windowSize.height, scaleFactor, step, subsetSize, objects);\r
- std::cout << res << "Results: " << cv::Mat(objects).row(0).colRange(0, res) << std::endl;\r
+ cv::gpu::device::lbp::classifyStump(stage_mat, stage_mat.cols / sizeof(Stage), nodes_mat, leaves_mat, subsets_mat, features_mat,\r
+ integral, processingRectSize.width, processingRectSize.height, windowSize.width, windowSize.height, scaleFactor, step, subsetSize, objects, dclassified);\r
}\r
+ cudaMemcpy(classified, dclassified, sizeof(int), cudaMemcpyDeviceToHost);\r
+ std::cout << *classified << "Results: " << cv::Mat(objects).row(0).colRange(0, *classified) << std::endl;\r
// TODO: reject levels\r
\r
return 0;\r
{
int y = threadIdx.x * scale;
int x = blockIdx.x * scale;
- *n = 0;
- int i = 0;
int current_node = 0;
int current_leave = 0;
current_leave += 2;
}
- i = s;
if (sum < stage.threshold)
return;
}
rect.z = roundf(clWidth);
rect.w = roundf(clHeight);
- int res = atomicInc(n, 1000);
+ int res = atomicInc(n, 100);
objects(0, res) = rect;
}
- int classifyStump(const DevMem2Db mstages, const int nstages, const DevMem2Di mnodes, const DevMem2Df mleaves, const DevMem2Di msubsets, const DevMem2Db mfeatures,
+ classifyStump(const DevMem2Db mstages, const int nstages, const DevMem2Di mnodes, const DevMem2Df mleaves, const DevMem2Di msubsets, const DevMem2Db mfeatures,
const DevMem2Di integral, const int workWidth, const int workHeight, const int clWidth, const int clHeight, float scale, int step, int subsetSize,
- DevMem2D_<int4> objects)
+ DevMem2D_<int4> objects, unsigned int* classified)
{
int blocks = ceilf(workHeight / (float)step);
int threads = ceilf(workWidth / (float)step);
- printf("blocks %d, threads %d\n", blocks, threads);
+ // printf("blocks %d, threads %d\n", blocks, threads);
Stage* stages = (Stage*)(mstages.ptr());
ClNode* nodes = (ClNode*)(mnodes.ptr());
const float* leaves = mleaves.ptr();
const int* subsets = msubsets.ptr();
const uchar4* features = (uchar4*)(mfeatures.ptr());
- unsigned int * n, *h_n = new unsigned int[1];
- cudaMalloc(&n, sizeof(int));
+
lbp_classify_stump<<<blocks, threads>>>(stages, nstages, nodes, leaves, subsets, features, integral,
- workWidth, workHeight, clWidth, clHeight, scale, step, subsetSize, objects, n);
- cudaMemcpy(h_n, n, sizeof(int), cudaMemcpyDeviceToHost);
- return *h_n;
+ workWidth, workHeight, clWidth, clHeight, scale, step, subsetSize, objects, classified);
}
}
}}}
\ No newline at end of file