From 57fa745be879e9fcbe007651bd190d3a40de3f38 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Fri, 04 Aug 2017 20:31:12 +0000
Subject: [PATCH] Fixed speed of SO/DLL
---
src/yolo_v2_class.cpp | 33 ++++++++++++++++++++++++++++++++-
1 files changed, 32 insertions(+), 1 deletions(-)
diff --git a/src/yolo_v2_class.cpp b/src/yolo_v2_class.cpp
index d79021a..813a24f 100644
--- a/src/yolo_v2_class.cpp
+++ b/src/yolo_v2_class.cpp
@@ -35,12 +35,16 @@
YOLODLL_API Detector::Detector(std::string cfg_filename, std::string weight_filename, int gpu_id)
{
int old_gpu_index;
+#ifdef GPU
cudaGetDevice(&old_gpu_index);
+#endif
detector_gpu_ptr = std::make_shared<detector_gpu_t>();
detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
+#ifdef GPU
cudaSetDevice(gpu_id);
+#endif
network &net = detector_gpu.net;
net.gpu_index = gpu_id;
//gpu_index = i;
@@ -66,7 +70,9 @@
detector_gpu.probs = (float **)calloc(l.w*l.h*l.n, sizeof(float *));
for (j = 0; j < l.w*l.h*l.n; ++j) detector_gpu.probs[j] = (float *)calloc(l.classes, sizeof(float));
+#ifdef GPU
cudaSetDevice(old_gpu_index);
+#endif
}
@@ -84,12 +90,25 @@
free(detector_gpu.probs);
int old_gpu_index;
+#ifdef GPU
cudaGetDevice(&old_gpu_index);
cudaSetDevice(detector_gpu.net.gpu_index);
+#endif
free_network(detector_gpu.net);
+#ifdef GPU
cudaSetDevice(old_gpu_index);
+#endif
+}
+
+YOLODLL_API int Detector::get_net_width() {
+ detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
+ return detector_gpu.net.w;
+}
+YOLODLL_API int Detector::get_net_height() {
+ detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
+ return detector_gpu.net.h;
}
@@ -150,8 +169,10 @@
detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
network &net = detector_gpu.net;
int old_gpu_index;
+#ifdef GPU
cudaGetDevice(&old_gpu_index);
cudaSetDevice(net.gpu_index);
+#endif
//std::cout << "net.gpu_index = " << net.gpu_index << std::endl;
//float nms = .4;
@@ -162,7 +183,15 @@
im.h = img.h;
im.w = img.w;
- image sized = resize_image(im, net.w, net.h);
+ image sized;
+
+ if (net.w == im.w && net.h == im.h) {
+ sized = make_image(im.w, im.h, im.c);
+ memcpy(sized.data, im.data, im.w*im.h*im.c * sizeof(float));
+ }
+ else
+ sized = resize_image(im, net.w, net.h);
+
layer l = net.layers[net.n - 1];
float *X = sized.data;
@@ -198,7 +227,9 @@
if(sized.data)
free(sized.data);
+#ifdef GPU
cudaSetDevice(old_gpu_index);
+#endif
return bbox_vec;
}
--
Gitblit v1.10.0