From 0f645836f193e75c4c3b718369e6fab15b5d19c5 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Wed, 11 Feb 2015 03:41:03 +0000
Subject: [PATCH] Detection is back, baby\!
---
src/parser.c | 241 +++++++++++++++++++++++++++++++++++++++++++++---
1 files changed, 226 insertions(+), 15 deletions(-)
diff --git a/src/parser.c b/src/parser.c
index 37ceb08..3f94c80 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -7,6 +7,7 @@
#include "crop_layer.h"
#include "cost_layer.h"
#include "convolutional_layer.h"
+#include "deconvolutional_layer.h"
#include "connected_layer.h"
#include "maxpool_layer.h"
#include "normalization_layer.h"
@@ -23,6 +24,7 @@
}section;
int is_convolutional(section *s);
+int is_deconvolutional(section *s);
int is_connected(section *s);
int is_maxpool(section *s);
int is_dropout(section *s);
@@ -65,6 +67,49 @@
}
}
+deconvolutional_layer *parse_deconvolutional(list *options, network *net, int count)
+{
+ int h,w,c;
+ float learning_rate, momentum, decay;
+ int n = option_find_int(options, "filters",1);
+ int size = option_find_int(options, "size",1);
+ int stride = option_find_int(options, "stride",1);
+ char *activation_s = option_find_str(options, "activation", "sigmoid");
+ ACTIVATION activation = get_activation(activation_s);
+ if(count == 0){
+ learning_rate = option_find_float(options, "learning_rate", .001);
+ momentum = option_find_float(options, "momentum", .9);
+ decay = option_find_float(options, "decay", .0001);
+ h = option_find_int(options, "height",1);
+ w = option_find_int(options, "width",1);
+ c = option_find_int(options, "channels",1);
+ net->batch = option_find_int(options, "batch",1);
+ net->learning_rate = learning_rate;
+ net->momentum = momentum;
+ net->decay = decay;
+ net->seen = option_find_int(options, "seen",0);
+ }else{
+ learning_rate = option_find_float_quiet(options, "learning_rate", net->learning_rate);
+ momentum = option_find_float_quiet(options, "momentum", net->momentum);
+ decay = option_find_float_quiet(options, "decay", net->decay);
+ image m = get_network_image_layer(*net, count-1);
+ h = m.h;
+ w = m.w;
+ c = m.c;
+ if(h == 0) error("Layer before deconvolutional layer must output image.");
+ }
+ deconvolutional_layer *layer = make_deconvolutional_layer(net->batch,h,w,c,n,size,stride,activation,learning_rate,momentum,decay);
+ char *weights = option_find_str(options, "weights", 0);
+ char *biases = option_find_str(options, "biases", 0);
+ parse_data(weights, layer->filters, c*n*size*size);
+ parse_data(biases, layer->biases, n);
+ #ifdef GPU
+ if(weights || biases) push_deconvolutional_layer(*layer);
+ #endif
+ option_unused(options);
+ return layer;
+}
+
convolutional_layer *parse_convolutional(list *options, network *net, int count)
{
int h,w,c;
@@ -86,6 +131,7 @@
net->learning_rate = learning_rate;
net->momentum = momentum;
net->decay = decay;
+ net->seen = option_find_int(options, "seen",0);
}else{
learning_rate = option_find_float_quiet(options, "learning_rate", net->learning_rate);
momentum = option_find_float_quiet(options, "momentum", net->momentum);
@@ -102,7 +148,7 @@
parse_data(weights, layer->filters, c*n*size*size);
parse_data(biases, layer->biases, n);
#ifdef GPU
- push_convolutional_layer(*layer);
+ if(weights || biases) push_convolutional_layer(*layer);
#endif
option_unused(options);
return layer;
@@ -136,7 +182,7 @@
parse_data(biases, layer->biases, output);
parse_data(weights, layer->weights, input*output);
#ifdef GPU
- push_connected_layer(*layer);
+ if(weights || biases) push_connected_layer(*layer);
#endif
option_unused(options);
return layer;
@@ -148,6 +194,7 @@
if(count == 0){
input = option_find_int(options, "input",1);
net->batch = option_find_int(options, "batch",1);
+ net->seen = option_find_int(options, "seen",0);
}else{
input = get_network_output_size_layer(*net, count-1);
}
@@ -162,6 +209,7 @@
if(count == 0){
input = option_find_int(options, "input",1);
net->batch = option_find_int(options, "batch",1);
+ net->seen = option_find_int(options, "seen",0);
}else{
input = get_network_output_size_layer(*net, count-1);
}
@@ -190,6 +238,7 @@
net->learning_rate = learning_rate;
net->momentum = momentum;
net->decay = decay;
+ net->seen = option_find_int(options, "seen",0);
}else{
image m = get_network_image_layer(*net, count-1);
h = m.h;
@@ -212,6 +261,7 @@
w = option_find_int(options, "width",1);
c = option_find_int(options, "channels",1);
net->batch = option_find_int(options, "batch",1);
+ net->seen = option_find_int(options, "seen",0);
}else{
image m = get_network_image_layer(*net, count-1);
h = m.h;
@@ -224,6 +274,7 @@
return layer;
}
+/*
freeweight_layer *parse_freeweight(list *options, network *net, int count)
{
int input;
@@ -237,6 +288,7 @@
option_unused(options);
return layer;
}
+*/
dropout_layer *parse_dropout(list *options, network *net, int count)
{
@@ -251,6 +303,7 @@
net->learning_rate = learning_rate;
net->momentum = momentum;
net->decay = decay;
+ net->seen = option_find_int(options, "seen",0);
}else{
input = get_network_output_size_layer(*net, count-1);
}
@@ -271,6 +324,7 @@
w = option_find_int(options, "width",1);
c = option_find_int(options, "channels",1);
net->batch = option_find_int(options, "batch",1);
+ net->seen = option_find_int(options, "seen",0);
}else{
image m = get_network_image_layer(*net, count-1);
h = m.h;
@@ -297,6 +351,10 @@
convolutional_layer *layer = parse_convolutional(options, &net, count);
net.types[count] = CONVOLUTIONAL;
net.layers[count] = layer;
+ }else if(is_deconvolutional(s)){
+ deconvolutional_layer *layer = parse_deconvolutional(options, &net, count);
+ net.types[count] = DECONVOLUTIONAL;
+ net.layers[count] = layer;
}else if(is_connected(s)){
connected_layer *layer = parse_connected(options, &net, count);
net.types[count] = CONNECTED;
@@ -326,9 +384,10 @@
net.types[count] = DROPOUT;
net.layers[count] = layer;
}else if(is_freeweight(s)){
- freeweight_layer *layer = parse_freeweight(options, &net, count);
- net.types[count] = FREEWEIGHT;
- net.layers[count] = layer;
+ //freeweight_layer *layer = parse_freeweight(options, &net, count);
+ //net.types[count] = FREEWEIGHT;
+ //net.layers[count] = layer;
+ fprintf(stderr, "Type not recognized: %s\n", s->type);
}else{
fprintf(stderr, "Type not recognized: %s\n", s->type);
}
@@ -350,6 +409,11 @@
{
return (strcmp(s->type, "[cost]")==0);
}
+int is_deconvolutional(section *s)
+{
+ return (strcmp(s->type, "[deconv]")==0
+ || strcmp(s->type, "[deconvolutional]")==0);
+}
int is_convolutional(section *s)
{
return (strcmp(s->type, "[conv]")==0
@@ -387,8 +451,8 @@
int read_option(char *s, list *options)
{
- int i;
- int len = strlen(s);
+ size_t i;
+ size_t len = strlen(s);
char *val = 0;
for(i = 0; i < len; ++i){
if(s[i] == '='){
@@ -416,7 +480,6 @@
strip(line);
switch(line[0]){
case '[':
- printf("%s\n", line);
current = malloc(sizeof(section));
list_insert(sections, current);
current->options = make_list();
@@ -429,7 +492,7 @@
break;
default:
if(!read_option(line, current->options)){
- printf("Config file error line %d, could parse: %s\n", nu, line);
+ fprintf(stderr, "Config file error line %d, could parse: %s\n", nu, line);
free(line);
}
break;
@@ -441,6 +504,9 @@
void print_convolutional_cfg(FILE *fp, convolutional_layer *l, network net, int count)
{
+ #ifdef GPU
+ if(gpu_index >= 0) pull_convolutional_layer(*l);
+ #endif
int i;
fprintf(fp, "[convolutional]\n");
if(count == 0) {
@@ -450,8 +516,9 @@
"channels=%d\n"
"learning_rate=%g\n"
"momentum=%g\n"
- "decay=%g\n",
- l->batch,l->h, l->w, l->c, l->learning_rate, l->momentum, l->decay);
+ "decay=%g\n"
+ "seen=%d\n",
+ l->batch,l->h, l->w, l->c, l->learning_rate, l->momentum, l->decay, net.seen);
} else {
if(l->learning_rate != net.learning_rate)
fprintf(fp, "learning_rate=%g\n", l->learning_rate);
@@ -475,6 +542,45 @@
fprintf(fp, "\n\n");
}
+void print_deconvolutional_cfg(FILE *fp, deconvolutional_layer *l, network net, int count)
+{
+ #ifdef GPU
+ if(gpu_index >= 0) pull_deconvolutional_layer(*l);
+ #endif
+ int i;
+ fprintf(fp, "[deconvolutional]\n");
+ if(count == 0) {
+ fprintf(fp, "batch=%d\n"
+ "height=%d\n"
+ "width=%d\n"
+ "channels=%d\n"
+ "learning_rate=%g\n"
+ "momentum=%g\n"
+ "decay=%g\n"
+ "seen=%d\n",
+ l->batch,l->h, l->w, l->c, l->learning_rate, l->momentum, l->decay, net.seen);
+ } else {
+ if(l->learning_rate != net.learning_rate)
+ fprintf(fp, "learning_rate=%g\n", l->learning_rate);
+ if(l->momentum != net.momentum)
+ fprintf(fp, "momentum=%g\n", l->momentum);
+ if(l->decay != net.decay)
+ fprintf(fp, "decay=%g\n", l->decay);
+ }
+ fprintf(fp, "filters=%d\n"
+ "size=%d\n"
+ "stride=%d\n"
+ "activation=%s\n",
+ l->n, l->size, l->stride,
+ get_activation_string(l->activation));
+ fprintf(fp, "biases=");
+ for(i = 0; i < l->n; ++i) fprintf(fp, "%g,", l->biases[i]);
+ fprintf(fp, "\n");
+ fprintf(fp, "weights=");
+ for(i = 0; i < l->n*l->c*l->size*l->size; ++i) fprintf(fp, "%g,", l->filters[i]);
+ fprintf(fp, "\n\n");
+}
+
void print_freeweight_cfg(FILE *fp, freeweight_layer *l, network net, int count)
{
fprintf(fp, "[freeweight]\n");
@@ -495,6 +601,9 @@
void print_connected_cfg(FILE *fp, connected_layer *l, network net, int count)
{
+ #ifdef GPU
+ if(gpu_index >= 0) pull_connected_layer(*l);
+ #endif
int i;
fprintf(fp, "[connected]\n");
if(count == 0){
@@ -502,8 +611,9 @@
"input=%d\n"
"learning_rate=%g\n"
"momentum=%g\n"
- "decay=%g\n",
- l->batch, l->inputs, l->learning_rate, l->momentum, l->decay);
+ "decay=%g\n"
+ "seen=%d\n",
+ l->batch, l->inputs, l->learning_rate, l->momentum, l->decay, net.seen);
} else {
if(l->learning_rate != net.learning_rate)
fprintf(fp, "learning_rate=%g\n", l->learning_rate);
@@ -534,8 +644,9 @@
"channels=%d\n"
"learning_rate=%g\n"
"momentum=%g\n"
- "decay=%g\n",
- l->batch,l->h, l->w, l->c, net.learning_rate, net.momentum, net.decay);
+ "decay=%g\n"
+ "seen=%d\n",
+ l->batch,l->h, l->w, l->c, net.learning_rate, net.momentum, net.decay, net.seen);
}
fprintf(fp, "crop_height=%d\ncrop_width=%d\nflip=%d\n\n", l->crop_height, l->crop_width, l->flip);
}
@@ -579,6 +690,104 @@
fprintf(fp, "\n");
}
+void save_weights(network net, char *filename)
+{
+ fprintf(stderr, "Saving weights to %s\n", filename);
+ FILE *fp = fopen(filename, "w");
+ if(!fp) file_error(filename);
+
+ fwrite(&net.learning_rate, sizeof(float), 1, fp);
+ fwrite(&net.momentum, sizeof(float), 1, fp);
+ fwrite(&net.decay, sizeof(float), 1, fp);
+ fwrite(&net.seen, sizeof(int), 1, fp);
+
+ int i;
+ for(i = 0; i < net.n; ++i){
+ if(net.types[i] == CONVOLUTIONAL){
+ convolutional_layer layer = *(convolutional_layer *) net.layers[i];
+ #ifdef GPU
+ if(gpu_index >= 0){
+ pull_convolutional_layer(layer);
+ }
+ #endif
+ int num = layer.n*layer.c*layer.size*layer.size;
+ fwrite(layer.biases, sizeof(float), layer.n, fp);
+ fwrite(layer.filters, sizeof(float), num, fp);
+ }
+ if(net.types[i] == DECONVOLUTIONAL){
+ deconvolutional_layer layer = *(deconvolutional_layer *) net.layers[i];
+ #ifdef GPU
+ if(gpu_index >= 0){
+ pull_deconvolutional_layer(layer);
+ }
+ #endif
+ int num = layer.n*layer.c*layer.size*layer.size;
+ fwrite(layer.biases, sizeof(float), layer.n, fp);
+ fwrite(layer.filters, sizeof(float), num, fp);
+ }
+ if(net.types[i] == CONNECTED){
+ connected_layer layer = *(connected_layer *) net.layers[i];
+ #ifdef GPU
+ if(gpu_index >= 0){
+ pull_connected_layer(layer);
+ }
+ #endif
+ fwrite(layer.biases, sizeof(float), layer.outputs, fp);
+ fwrite(layer.weights, sizeof(float), layer.outputs*layer.inputs, fp);
+ }
+ }
+ fclose(fp);
+}
+
+void load_weights(network *net, char *filename)
+{
+ fprintf(stderr, "Loading weights from %s\n", filename);
+ FILE *fp = fopen(filename, "r");
+ if(!fp) file_error(filename);
+
+ fread(&net->learning_rate, sizeof(float), 1, fp);
+ fread(&net->momentum, sizeof(float), 1, fp);
+ fread(&net->decay, sizeof(float), 1, fp);
+ fread(&net->seen, sizeof(int), 1, fp);
+ set_learning_network(net, net->learning_rate, net->momentum, net->decay);
+
+ int i;
+ for(i = 0; i < net->n; ++i){
+ if(net->types[i] == CONVOLUTIONAL){
+ convolutional_layer layer = *(convolutional_layer *) net->layers[i];
+ int num = layer.n*layer.c*layer.size*layer.size;
+ fread(layer.biases, sizeof(float), layer.n, fp);
+ fread(layer.filters, sizeof(float), num, fp);
+ #ifdef GPU
+ if(gpu_index >= 0){
+ push_convolutional_layer(layer);
+ }
+ #endif
+ }
+ if(net->types[i] == DECONVOLUTIONAL){
+ deconvolutional_layer layer = *(deconvolutional_layer *) net->layers[i];
+ int num = layer.n*layer.c*layer.size*layer.size;
+ fread(layer.biases, sizeof(float), layer.n, fp);
+ fread(layer.filters, sizeof(float), num, fp);
+ #ifdef GPU
+ if(gpu_index >= 0){
+ push_deconvolutional_layer(layer);
+ }
+ #endif
+ }
+ if(net->types[i] == CONNECTED){
+ connected_layer layer = *(connected_layer *) net->layers[i];
+ fread(layer.biases, sizeof(float), layer.outputs, fp);
+ fread(layer.weights, sizeof(float), layer.outputs*layer.inputs, fp);
+ #ifdef GPU
+ if(gpu_index >= 0){
+ push_connected_layer(layer);
+ }
+ #endif
+ }
+ }
+ fclose(fp);
+}
void save_network(network net, char *filename)
{
@@ -589,6 +798,8 @@
{
if(net.types[i] == CONVOLUTIONAL)
print_convolutional_cfg(fp, (convolutional_layer *)net.layers[i], net, i);
+ else if(net.types[i] == DECONVOLUTIONAL)
+ print_deconvolutional_cfg(fp, (deconvolutional_layer *)net.layers[i], net, i);
else if(net.types[i] == CONNECTED)
print_connected_cfg(fp, (connected_layer *)net.layers[i], net, i);
else if(net.types[i] == CROP)
--
Gitblit v1.10.0