From bc810016a1f2eadf33a1ac800b64962a42f3d402 Mon Sep 17 00:00:00 2001 From: AlexeyAB Date: Thu, 3 Aug 2017 01:36:22 +0300 Subject: [PATCH] cuDNN 6.0 supported. Also speed of console example improved. --- src/convolutional_layer.c | 8 ++++++-- src/yolo_console_dll.cpp | 19 +++++++++++++------ 2 files changed, 19 insertions(+), 8 deletions(-) diff --git a/src/convolutional_layer.c b/src/convolutional_layer.c index cf5d2529..a3247d03 100644 --- a/src/convolutional_layer.c +++ b/src/convolutional_layer.c @@ -146,8 +146,12 @@ void cudnn_convolutional_setup(layer *l) cudnnSetTensor4dDescriptor(l->srcTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->c, l->h, l->w); cudnnSetTensor4dDescriptor(l->dstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->out_c, l->out_h, l->out_w); cudnnSetFilter4dDescriptor(l->weightDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c, l->size, l->size); - cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION); - cudnnGetConvolutionForwardAlgorithm(cudnn_handle(), +#if(CUDNN_MAJOR >= 6) + cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION, CUDNN_DATA_FLOAT); // cudnn 6.0 +#else + cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION); // cudnn 5.1 +#endif + cudnnGetConvolutionForwardAlgorithm(cudnn_handle(), l->srcTensorDesc, l->weightDesc, l->convDesc, diff --git a/src/yolo_console_dll.cpp b/src/yolo_console_dll.cpp index dbdaf85f..adbe873b 100644 --- a/src/yolo_console_dll.cpp +++ b/src/yolo_console_dll.cpp @@ -73,14 +73,21 @@ int main() #ifdef OPENCV std::string const file_ext = filename.substr(filename.find_last_of(".") + 1); if (file_ext == "avi" || file_ext == "mp4" || file_ext == "mjpg" || file_ext == "mov") { // video file - cv::Mat frame; + cv::Mat frame, prev_frame; + std::vector result_vec, thread_result_vec; detector.nms = 0.02; // comment it - if track_id is not required - for(cv::VideoCapture cap(filename); cap >> frame, cap.isOpened();) { - std::vector result_vec = detector.detect(frame, 0.2); - result_vec = detector.tracking(result_vec); // comment it - if track_id is not required + for (cv::VideoCapture cap(filename); cap >> frame, cap.isOpened();) { + auto image_ptr = detector.mat_to_image(frame); + std::thread td([&]() { thread_result_vec = detector.detect(*image_ptr, 0.2); }); - draw_boxes(frame, result_vec, obj_names, 3); - show_result(result_vec, obj_names); + if (!prev_frame.empty()) { + result_vec = detector.tracking(result_vec); // comment it - if track_id is not required + draw_boxes(prev_frame, result_vec, obj_names, 3); + show_result(result_vec, obj_names); + } + td.join(); + prev_frame = frame; + result_vec = thread_result_vec; } } else { // image file