From a6c51e3b758aee7fd3a6f1d37daa8dcad4891e52 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Thu, 08 Mar 2018 22:42:58 +0000
Subject: [PATCH] Fixes for small objects. And fixes of densenet201_yolo.cfg, resnet50_yolo.cfg, resnet152_yolo.cfg

---
 src/network.c |   19 ++++++++++++++++---
 1 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/src/network.c b/src/network.c
index f3185a3..d23468d 100644
--- a/src/network.c
+++ b/src/network.c
@@ -316,7 +316,20 @@
         net->layers[i].batch = b;
 #ifdef CUDNN
         if(net->layers[i].type == CONVOLUTIONAL){
-            cudnn_convolutional_setup(net->layers + i);
+			cudnn_convolutional_setup(net->layers + i, cudnn_fastest);
+			/*
+			layer *l = net->layers + i;
+            cudnn_convolutional_setup(l, cudnn_fastest);
+			// check for excessive memory consumption 
+			size_t free_byte;
+			size_t total_byte;
+			check_error(cudaMemGetInfo(&free_byte, &total_byte));
+			if (l->workspace_size > free_byte || l->workspace_size >= total_byte / 2) {
+				printf(" used slow CUDNN algo without Workspace! \n");
+				cudnn_convolutional_setup(l, cudnn_smallest);
+				l->workspace_size = get_workspace_size(*l);
+			}
+			*/
         }
 #endif
     }
@@ -346,7 +359,7 @@
     //fflush(stderr);
     for (i = 0; i < net->n; ++i){
         layer l = net->layers[i];
-		printf(" %d: layer = %d,", i, l.type);
+		//printf(" %d: layer = %d,", i, l.type);
         if(l.type == CONVOLUTIONAL){
             resize_convolutional_layer(&l, w, h);
         }else if(l.type == CROP){
@@ -378,7 +391,7 @@
     }
 #ifdef GPU
     if(gpu_index >= 0){
-		printf(" try to allocate workspace, ");
+		printf(" try to allocate workspace = %zu * sizeof(float), ", (workspace_size - 1) / sizeof(float) + 1);
         net->workspace = cuda_make_array(0, (workspace_size-1)/sizeof(float)+1);
 		printf(" CUDA allocate done! \n");
     }else {

--
Gitblit v1.10.0