From db0397cfaaf488364e3d2e1669dfefae2ee6ea73 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 14 Dec 2015 19:57:10 +0000
Subject: [PATCH] shortcut layers, msr networks
---
src/parser.c | 118 ++++++++++++++++++++++++++++++++++++++++++++++++++++++----
1 files changed, 109 insertions(+), 9 deletions(-)
diff --git a/src/parser.c b/src/parser.c
index b095294..8efafad 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -15,7 +15,9 @@
#include "dropout_layer.h"
#include "detection_layer.h"
#include "avgpool_layer.h"
+#include "local_layer.h"
#include "route_layer.h"
+#include "shortcut_layer.h"
#include "list.h"
#include "option_list.h"
#include "utils.h"
@@ -27,6 +29,7 @@
int is_network(section *s);
int is_convolutional(section *s);
+int is_local(section *s);
int is_deconvolutional(section *s);
int is_connected(section *s);
int is_maxpool(section *s);
@@ -35,6 +38,7 @@
int is_softmax(section *s);
int is_normalization(section *s);
int is_crop(section *s);
+int is_shortcut(section *s);
int is_cost(section *s);
int is_detection(section *s);
int is_route(section *s);
@@ -78,6 +82,7 @@
int h;
int w;
int c;
+ int index;
} size_params;
deconvolutional_layer parse_deconvolutional(list *options, size_params params)
@@ -107,6 +112,27 @@
return layer;
}
+local_layer parse_local(list *options, size_params params)
+{
+ int n = option_find_int(options, "filters",1);
+ int size = option_find_int(options, "size",1);
+ int stride = option_find_int(options, "stride",1);
+ int pad = option_find_int(options, "pad",0);
+ char *activation_s = option_find_str(options, "activation", "logistic");
+ ACTIVATION activation = get_activation(activation_s);
+
+ int batch,h,w,c;
+ h = params.h;
+ w = params.w;
+ c = params.c;
+ batch=params.batch;
+ if(!(h && w && c)) error("Layer before local layer must output image.");
+
+ local_layer layer = make_local_layer(batch,h,w,c,n,size,stride,pad,activation);
+
+ return layer;
+}
+
convolutional_layer parse_convolutional(list *options, size_params params)
{
int n = option_find_int(options, "filters",1);
@@ -125,6 +151,7 @@
int batch_normalize = option_find_int_quiet(options, "batch_normalize", 0);
convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,size,stride,pad,activation, batch_normalize);
+ layer.flipped = option_find_int_quiet(options, "flipped", 0);
char *weights = option_find_str(options, "weights", 0);
char *biases = option_find_str(options, "biases", 0);
@@ -264,6 +291,20 @@
return l;
}
+layer parse_shortcut(list *options, size_params params, network net)
+{
+ char *l = option_find(options, "from");
+ int index = atoi(l);
+ if(index < 0) index = params.index + index;
+
+ int batch = params.batch;
+ layer from = net.layers[index];
+
+ layer s = make_shortcut_layer(batch, index, params.w, params.h, params.c, from.out_w, from.out_h, from.out_c);
+ return s;
+}
+
+
route_layer parse_route(list *options, size_params params, network net)
{
char *l = option_find(options, "layers");
@@ -280,13 +321,14 @@
for(i = 0; i < n; ++i){
int index = atoi(l);
l = strchr(l, ',')+1;
+ if(index < 0) index = params.index + index;
layers[i] = index;
sizes[i] = net.layers[index].outputs;
}
int batch = params.batch;
route_layer layer = make_route_layer(batch, n, layers, sizes);
-
+
convolutional_layer first = net.layers[layers[0]];
layer.out_w = first.out_w;
layer.out_h = first.out_h;
@@ -396,12 +438,15 @@
int count = 0;
free_section(s);
while(n){
+ params.index = count;
fprintf(stderr, "%d: ", count);
s = (section *)n->val;
options = s->options;
layer l = {0};
if(is_convolutional(s)){
l = parse_convolutional(options, params);
+ }else if(is_local(s)){
+ l = parse_local(options, params);
}else if(is_deconvolutional(s)){
l = parse_deconvolutional(options, params);
}else if(is_connected(s)){
@@ -422,6 +467,8 @@
l = parse_avgpool(options, params);
}else if(is_route(s)){
l = parse_route(options, params, net);
+ }else if(is_shortcut(s)){
+ l = parse_shortcut(options, params, net);
}else if(is_dropout(s)){
l = parse_dropout(options, params);
l.output = net.layers[count-1].output;
@@ -439,13 +486,13 @@
net.layers[count] = l;
free_section(s);
n = n->next;
+ ++count;
if(n){
params.h = l.out_h;
params.w = l.out_w;
params.c = l.out_c;
params.inputs = l.outputs;
}
- ++count;
}
free_list(sections);
net.outputs = get_network_output_size(net);
@@ -453,6 +500,10 @@
return net;
}
+int is_shortcut(section *s)
+{
+ return (strcmp(s->type, "[shortcut]")==0);
+}
int is_crop(section *s)
{
return (strcmp(s->type, "[crop]")==0);
@@ -465,6 +516,10 @@
{
return (strcmp(s->type, "[detection]")==0);
}
+int is_local(section *s)
+{
+ return (strcmp(s->type, "[local]")==0);
+}
int is_deconvolutional(section *s)
{
return (strcmp(s->type, "[deconv]")==0
@@ -596,9 +651,12 @@
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);
+ int major = 0;
+ int minor = 1;
+ int revision = 0;
+ fwrite(&major, sizeof(int), 1, fp);
+ fwrite(&minor, sizeof(int), 1, fp);
+ fwrite(&revision, sizeof(int), 1, fp);
fwrite(net.seen, sizeof(int), 1, fp);
int i;
@@ -626,6 +684,16 @@
#endif
fwrite(l.biases, sizeof(float), l.outputs, fp);
fwrite(l.weights, sizeof(float), l.outputs*l.inputs, fp);
+ } if(l.type == LOCAL){
+#ifdef GPU
+ if(gpu_index >= 0){
+ pull_local_layer(l);
+ }
+#endif
+ int locations = l.out_w*l.out_h;
+ int size = l.size*l.size*l.c*l.n*locations;
+ fwrite(l.biases, sizeof(float), l.outputs, fp);
+ fwrite(l.filters, sizeof(float), size, fp);
}
}
fclose(fp);
@@ -635,6 +703,19 @@
save_weights_upto(net, filename, net.n);
}
+void transpose_matrix(float *a, int rows, int cols)
+{
+ float *transpose = calloc(rows*cols, sizeof(float));
+ int x, y;
+ for(x = 0; x < rows; ++x){
+ for(y = 0; y < cols; ++y){
+ transpose[y*rows + x] = a[x*cols + y];
+ }
+ }
+ memcpy(a, transpose, rows*cols*sizeof(float));
+ free(transpose);
+}
+
void load_weights_upto(network *net, char *filename, int cutoff)
{
fprintf(stderr, "Loading weights from %s...", filename);
@@ -642,10 +723,12 @@
FILE *fp = fopen(filename, "r");
if(!fp) file_error(filename);
- float garbage;
- fread(&garbage, sizeof(float), 1, fp);
- fread(&garbage, sizeof(float), 1, fp);
- fread(&garbage, sizeof(float), 1, fp);
+ int major;
+ int minor;
+ int revision;
+ fread(&major, sizeof(int), 1, fp);
+ fread(&minor, sizeof(int), 1, fp);
+ fread(&revision, sizeof(int), 1, fp);
fread(net->seen, sizeof(int), 1, fp);
int i;
@@ -661,6 +744,9 @@
fread(l.rolling_variance, sizeof(float), l.n, fp);
}
fread(l.filters, sizeof(float), num, fp);
+ if (l.flipped) {
+ transpose_matrix(l.filters, l.c*l.size*l.size, l.n);
+ }
#ifdef GPU
if(gpu_index >= 0){
push_convolutional_layer(l);
@@ -680,12 +766,26 @@
if(l.type == CONNECTED){
fread(l.biases, sizeof(float), l.outputs, fp);
fread(l.weights, sizeof(float), l.outputs*l.inputs, fp);
+ if(major > 1000 || minor > 1000){
+ transpose_matrix(l.weights, l.inputs, l.outputs);
+ }
#ifdef GPU
if(gpu_index >= 0){
push_connected_layer(l);
}
#endif
}
+ if(l.type == LOCAL){
+ int locations = l.out_w*l.out_h;
+ int size = l.size*l.size*l.c*l.n*locations;
+ fread(l.biases, sizeof(float), l.outputs, fp);
+ fread(l.filters, sizeof(float), size, fp);
+#ifdef GPU
+ if(gpu_index >= 0){
+ push_local_layer(l);
+ }
+#endif
+ }
}
fprintf(stderr, "Done!\n");
fclose(fp);
--
Gitblit v1.10.0