From 1a35f49ab3ae9d74b636de8580e07ab072846ea9 Mon Sep 17 00:00:00 2001
From: Alexey <AlexeyAB@users.noreply.github.com>
Date: Thu, 31 May 2018 11:19:42 +0000
Subject: [PATCH] Merge pull request #950 from IlyaOvodov/Fix_size_PR
---
src/network.c | 101 ++++++++++++++++++++++++++++++++++++++++++++++++--
1 files changed, 97 insertions(+), 4 deletions(-)
diff --git a/src/network.c b/src/network.c
index 8619158..050d334 100644
--- a/src/network.c
+++ b/src/network.c
@@ -28,6 +28,25 @@
#include "route_layer.h"
#include "shortcut_layer.h"
#include "yolo_layer.h"
+#include "upsample_layer.h"
+#include "parser.h"
+
+network *load_network_custom(char *cfg, char *weights, int clear, int batch)
+{
+ printf(" Try to load cfg: %s, weights: %s, clear = %d \n", cfg, weights, clear);
+ network *net = calloc(1, sizeof(network));
+ *net = parse_network_cfg_custom(cfg, batch);
+ if (weights && weights[0] != 0) {
+ load_weights(net, weights);
+ }
+ if (clear) (*net->seen) = 0;
+ return net;
+}
+
+network *load_network(char *cfg, char *weights, int clear)
+{
+ return load_network_custom(cfg, weights, clear, 0);
+}
int get_current_batch(network net)
{
@@ -46,6 +65,27 @@
#endif
}
+void reset_network_state(network *net, int b)
+{
+ int i;
+ for (i = 0; i < net->n; ++i) {
+#ifdef GPU
+ layer l = net->layers[i];
+ if (l.state_gpu) {
+ fill_ongpu(l.outputs, 0, l.state_gpu + l.outputs*b, 1);
+ }
+ if (l.h_gpu) {
+ fill_ongpu(l.outputs, 0, l.h_gpu + l.outputs*b, 1);
+ }
+#endif
+ }
+}
+
+void reset_rnn(network *net)
+{
+ reset_network_state(net, 0);
+}
+
float get_current_rate(network net)
{
int batch_num = get_current_batch(net);
@@ -138,7 +178,7 @@
net.n = n;
net.layers = calloc(net.n, sizeof(layer));
net.seen = calloc(1, sizeof(int));
- #ifdef GPU
+#ifdef GPU
net.input_gpu = calloc(1, sizeof(float *));
net.truth_gpu = calloc(1, sizeof(float *));
@@ -146,7 +186,7 @@
net.output16_gpu = calloc(1, sizeof(float *));
net.max_input16_size = calloc(1, sizeof(size_t));
net.max_output16_size = calloc(1, sizeof(size_t));
- #endif
+#endif
return net;
}
@@ -374,10 +414,14 @@
resize_maxpool_layer(&l, w, h);
}else if(l.type == REGION){
resize_region_layer(&l, w, h);
+ }else if (l.type == YOLO) {
+ resize_yolo_layer(&l, w, h);
}else if(l.type == ROUTE){
resize_route_layer(&l, net);
}else if (l.type == SHORTCUT) {
resize_shortcut_layer(&l, w, h);
+ }else if (l.type == UPSAMPLE) {
+ resize_upsample_layer(&l, w, h);
}else if(l.type == REORG){
resize_reorg_layer(&l, w, h);
}else if(l.type == AVGPOOL){
@@ -538,17 +582,22 @@
box *boxes = calloc(l.w*l.h*l.n, sizeof(box));
float **probs = calloc(l.w*l.h*l.n, sizeof(float *));
int i, j;
- for (j = 0; j < l.w*l.h*l.n; ++j) probs[j] = calloc(l.classes, sizeof(float *));
+ for (j = 0; j < l.w*l.h*l.n; ++j) probs[j] = calloc(l.classes, sizeof(float));
get_region_boxes(l, 1, 1, thresh, probs, boxes, 0, map);
for (j = 0; j < l.w*l.h*l.n; ++j) {
dets[j].classes = l.classes;
dets[j].bbox = boxes[j];
dets[j].objectness = 1;
- for (i = 0; i < l.classes; ++i) dets[j].prob[i] = probs[j][i];
+ for (i = 0; i < l.classes; ++i) {
+ dets[j].prob[i] = probs[j][i];
+ }
}
free(boxes);
free_ptrs((void **)probs, l.w*l.h*l.n);
+
+ //correct_region_boxes(dets, l.w*l.h*l.n, w, h, net_w, net_h, relative);
+ correct_yolo_boxes(dets, l.w*l.h*l.n, w, h, net_w, net_h, relative, letter);
}
void fill_network_boxes(network *net, int w, int h, float thresh, float hier, int *map, int relative, detection *dets, int letter)
@@ -724,6 +773,11 @@
free_layer(net.layers[i]);
}
free(net.layers);
+
+ free(net.scales);
+ free(net.steps);
+ free(net.seen);
+
#ifdef GPU
if (gpu_index >= 0) cuda_free(net.workspace);
else free(net.workspace);
@@ -742,3 +796,42 @@
free(net.workspace);
#endif
}
+
+
+void fuse_conv_batchnorm(network net)
+{
+ int j;
+ for (j = 0; j < net.n; ++j) {
+ layer *l = &net.layers[j];
+
+ if (l->type == CONVOLUTIONAL) {
+ //printf(" Merges Convolutional-%d and batch_norm \n", j);
+
+ if (l->batch_normalize) {
+ int f;
+ for (f = 0; f < l->n; ++f)
+ {
+ l->biases[f] = l->biases[f] - (double)l->scales[f] * l->rolling_mean[f] / (sqrt((double)l->rolling_variance[f]) + .000001f);
+
+ const size_t filter_size = l->size*l->size*l->c;
+ int i;
+ for (i = 0; i < filter_size; ++i) {
+ int w_index = f*filter_size + i;
+
+ l->weights[w_index] = (double)l->weights[w_index] * l->scales[f] / (sqrt((double)l->rolling_variance[f]) + .000001f);
+ }
+ }
+
+ l->batch_normalize = 0;
+#ifdef GPU
+ if (gpu_index >= 0) {
+ push_convolutional_layer(*l);
+ }
+#endif
+ }
+ }
+ else {
+ //printf(" Fusion skip layer type: %d \n", l->type);
+ }
+ }
+}
--
Gitblit v1.10.0