From 8c5364f58569eaeb5582a4915b36b24fc5570c76 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 09 Nov 2015 19:31:39 +0000
Subject: [PATCH] New YOLO

---
 src/parser.c |  242 +++++++++++++++++++++++++++++++++++++-----------
 1 files changed, 186 insertions(+), 56 deletions(-)

diff --git a/src/parser.c b/src/parser.c
index 240c6ee..b095294 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -7,12 +7,14 @@
 #include "crop_layer.h"
 #include "cost_layer.h"
 #include "convolutional_layer.h"
+#include "normalization_layer.h"
 #include "deconvolutional_layer.h"
 #include "connected_layer.h"
 #include "maxpool_layer.h"
 #include "softmax_layer.h"
 #include "dropout_layer.h"
 #include "detection_layer.h"
+#include "avgpool_layer.h"
 #include "route_layer.h"
 #include "list.h"
 #include "option_list.h"
@@ -28,8 +30,10 @@
 int is_deconvolutional(section *s);
 int is_connected(section *s);
 int is_maxpool(section *s);
+int is_avgpool(section *s);
 int is_dropout(section *s);
 int is_softmax(section *s);
+int is_normalization(section *s);
 int is_crop(section *s);
 int is_cost(section *s);
 int is_detection(section *s);
@@ -100,7 +104,6 @@
     #ifdef GPU
     if(weights || biases) push_deconvolutional_layer(layer);
     #endif
-    option_unused(options);
     return layer;
 }
 
@@ -119,8 +122,9 @@
     c = params.c;
     batch=params.batch;
     if(!(h && w && c)) error("Layer before convolutional layer must output image.");
+    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);
+    convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,size,stride,pad,activation, batch_normalize);
 
     char *weights = option_find_str(options, "weights", 0);
     char *biases = option_find_str(options, "biases", 0);
@@ -129,7 +133,6 @@
     #ifdef GPU
     if(weights || biases) push_convolutional_layer(layer);
     #endif
-    option_unused(options);
     return layer;
 }
 
@@ -148,7 +151,6 @@
     #ifdef GPU
     if(weights || biases) push_connected_layer(layer);
     #endif
-    option_unused(options);
     return layer;
 }
 
@@ -156,7 +158,6 @@
 {
     int groups = option_find_int(options, "groups",1);
     softmax_layer layer = make_softmax_layer(params.batch, params.inputs, groups);
-    option_unused(options);
     return layer;
 }
 
@@ -165,11 +166,19 @@
     int coords = option_find_int(options, "coords", 1);
     int classes = option_find_int(options, "classes", 1);
     int rescore = option_find_int(options, "rescore", 0);
-    int joint = option_find_int(options, "joint", 0);
-    int objectness = option_find_int(options, "objectness", 0);
-    int background = option_find_int(options, "background", 1);
-    detection_layer layer = make_detection_layer(params.batch, params.inputs, classes, coords, joint, rescore, background, objectness);
-    option_unused(options);
+    int num = option_find_int(options, "num", 1);
+    int side = option_find_int(options, "side", 7);
+    detection_layer layer = make_detection_layer(params.batch, params.inputs, num, side, classes, coords, rescore);
+
+    layer.softmax = option_find_int(options, "softmax", 0);
+    layer.sqrt = option_find_int(options, "sqrt", 0);
+
+    layer.coord_scale = option_find_float(options, "coord_scale", 1);
+    layer.forced = option_find_int(options, "forced", 0);
+    layer.object_scale = option_find_float(options, "object_scale", 1);
+    layer.noobject_scale = option_find_float(options, "noobject_scale", 1);
+    layer.class_scale = option_find_float(options, "class_scale", 1);
+    layer.jitter = option_find_float(options, "jitter", .2);
     return layer;
 }
 
@@ -177,8 +186,8 @@
 {
     char *type_s = option_find_str(options, "type", "sse");
     COST_TYPE type = get_cost_type(type_s);
-    cost_layer layer = make_cost_layer(params.batch, params.inputs, type);
-    option_unused(options);
+    float scale = option_find_float_quiet(options, "scale",1);
+    cost_layer layer = make_cost_layer(params.batch, params.inputs, type, scale);
     return layer;
 }
 
@@ -198,8 +207,11 @@
     batch=params.batch;
     if(!(h && w && c)) error("Layer before crop layer must output image.");
 
+    int noadjust = option_find_int_quiet(options, "noadjust",0);
+
     crop_layer l = make_crop_layer(batch,h,w,c,crop_height,crop_width,flip, angle, saturation, exposure);
-    option_unused(options);
+    l.shift = option_find_float(options, "shift", 0);
+    l.noadjust = noadjust;
     return l;
 }
 
@@ -216,7 +228,19 @@
     if(!(h && w && c)) error("Layer before maxpool layer must output image.");
 
     maxpool_layer layer = make_maxpool_layer(batch,h,w,c,size,stride);
-    option_unused(options);
+    return layer;
+}
+
+avgpool_layer parse_avgpool(list *options, size_params params)
+{
+    int batch,w,h,c;
+    w = params.w;
+    h = params.h;
+    c = params.c;
+    batch=params.batch;
+    if(!(h && w && c)) error("Layer before avgpool layer must output image.");
+
+    avgpool_layer layer = make_avgpool_layer(batch,w,h,c);
     return layer;
 }
 
@@ -224,10 +248,22 @@
 {
     float probability = option_find_float(options, "probability", .5);
     dropout_layer layer = make_dropout_layer(params.batch, params.inputs, probability);
-    option_unused(options);
+    layer.out_w = params.w;
+    layer.out_h = params.h;
+    layer.out_c = params.c;
     return layer;
 }
 
+layer parse_normalization(list *options, size_params params)
+{
+    float alpha = option_find_float(options, "alpha", .0001);
+    float beta =  option_find_float(options, "beta" , .75);
+    float kappa = option_find_float(options, "kappa", 1);
+    int size = option_find_int(options, "size", 5);
+    layer l = make_normalization_layer(params.batch, params.w, params.h, params.c, size, alpha, beta, kappa);
+    return l;
+}
+
 route_layer parse_route(list *options, size_params params, network net)
 {
     char *l = option_find(options, "layers");   
@@ -265,17 +301,27 @@
         }
     }
 
-    option_unused(options);
     return layer;
 }
 
+learning_rate_policy get_policy(char *s)
+{
+    if (strcmp(s, "poly")==0) return POLY;
+    if (strcmp(s, "constant")==0) return CONSTANT;
+    if (strcmp(s, "step")==0) return STEP;
+    if (strcmp(s, "exp")==0) return EXP;
+    if (strcmp(s, "sigmoid")==0) return SIG;
+    if (strcmp(s, "steps")==0) return STEPS;
+    fprintf(stderr, "Couldn't find policy %s, going with constant\n", s);
+    return CONSTANT;
+}
+
 void parse_net_options(list *options, network *net)
 {
     net->batch = option_find_int(options, "batch",1);
     net->learning_rate = option_find_float(options, "learning_rate", .001);
     net->momentum = option_find_float(options, "momentum", .9);
     net->decay = option_find_float(options, "decay", .0001);
-    net->seen = option_find_int(options, "seen",0);
     int subdivs = option_find_int(options, "subdivisions",1);
     net->batch /= subdivs;
     net->subdivisions = subdivs;
@@ -284,8 +330,47 @@
     net->w = option_find_int_quiet(options, "width",0);
     net->c = option_find_int_quiet(options, "channels",0);
     net->inputs = option_find_int_quiet(options, "inputs", net->h * net->w * net->c);
+
     if(!net->inputs && !(net->h && net->w && net->c)) error("No input parameters supplied");
-    option_unused(options);
+
+    char *policy_s = option_find_str(options, "policy", "constant");
+    net->policy = get_policy(policy_s);
+    if(net->policy == STEP){
+        net->step = option_find_int(options, "step", 1);
+        net->scale = option_find_float(options, "scale", 1);
+    } else if (net->policy == STEPS){
+        char *l = option_find(options, "steps");   
+        char *p = option_find(options, "scales");   
+        if(!l || !p) error("STEPS policy must have steps and scales in cfg file");
+
+        int len = strlen(l);
+        int n = 1;
+        int i;
+        for(i = 0; i < len; ++i){
+            if (l[i] == ',') ++n;
+        }
+        int *steps = calloc(n, sizeof(int));
+        float *scales = calloc(n, sizeof(float));
+        for(i = 0; i < n; ++i){
+            int step    = atoi(l);
+            float scale = atof(p);
+            l = strchr(l, ',')+1;
+            p = strchr(p, ',')+1;
+            steps[i] = step;
+            scales[i] = scale;
+        }
+        net->scales = scales;
+        net->steps = steps;
+        net->num_steps = n;
+    } else if (net->policy == EXP){
+        net->gamma = option_find_float(options, "gamma", 1);
+    } else if (net->policy == SIG){
+        net->gamma = option_find_float(options, "gamma", 1);
+        net->step = option_find_int(options, "step", 1);
+    } else if (net->policy == POLY){
+        net->power = option_find_float(options, "power", 1);
+    }
+    net->max_batches = option_find_int(options, "max_batches", 0);
 }
 
 network parse_network_cfg(char *filename)
@@ -309,6 +394,7 @@
 
     n = n->next;
     int count = 0;
+    free_section(s);
     while(n){
         fprintf(stderr, "%d: ", count);
         s = (section *)n->val;
@@ -328,21 +414,28 @@
             l = parse_detection(options, params);
         }else if(is_softmax(s)){
             l = parse_softmax(options, params);
+        }else if(is_normalization(s)){
+            l = parse_normalization(options, params);
         }else if(is_maxpool(s)){
             l = parse_maxpool(options, params);
+        }else if(is_avgpool(s)){
+            l = parse_avgpool(options, params);
         }else if(is_route(s)){
             l = parse_route(options, params, net);
         }else if(is_dropout(s)){
             l = parse_dropout(options, params);
             l.output = net.layers[count-1].output;
             l.delta = net.layers[count-1].delta;
-            #ifdef GPU
+#ifdef GPU
             l.output_gpu = net.layers[count-1].output_gpu;
             l.delta_gpu = net.layers[count-1].delta_gpu;
-            #endif
+#endif
         }else{
             fprintf(stderr, "Type not recognized: %s\n", s->type);
         }
+        l.dontload = option_find_int_quiet(options, "dontload", 0);
+        l.dontloadscales = option_find_int_quiet(options, "dontloadscales", 0);
+        option_unused(options);
         net.layers[count] = l;
         free_section(s);
         n = n->next;
@@ -397,11 +490,22 @@
     return (strcmp(s->type, "[max]")==0
             || strcmp(s->type, "[maxpool]")==0);
 }
+int is_avgpool(section *s)
+{
+    return (strcmp(s->type, "[avg]")==0
+            || strcmp(s->type, "[avgpool]")==0);
+}
 int is_dropout(section *s)
 {
     return (strcmp(s->type, "[dropout]")==0);
 }
 
+int is_normalization(section *s)
+{
+    return (strcmp(s->type, "[lrn]")==0
+            || strcmp(s->type, "[normalization]")==0);
+}
+
 int is_softmax(section *s)
 {
     return (strcmp(s->type, "[soft]")==0
@@ -412,24 +516,6 @@
     return (strcmp(s->type, "[route]")==0);
 }
 
-int read_option(char *s, list *options)
-{
-    size_t i;
-    size_t len = strlen(s);
-    char *val = 0;
-    for(i = 0; i < len; ++i){
-        if(s[i] == '='){
-            s[i] = '\0';
-            val = s+i+1;
-            break;
-        }
-    }
-    if(i == len-1) return 0;
-    char *key = s;
-    option_insert(options, key, val);
-    return 1;
-}
-
 list *read_cfg(char *filename)
 {
     FILE *file = fopen(filename, "r");
@@ -465,7 +551,46 @@
     return sections;
 }
 
-void save_weights(network net, char *filename)
+void save_weights_double(network net, char *filename)
+{
+    fprintf(stderr, "Saving doubled 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,j,k;
+    for(i = 0; i < net.n; ++i){
+        layer l = net.layers[i];
+        if(l.type == CONVOLUTIONAL){
+#ifdef GPU
+            if(gpu_index >= 0){
+                pull_convolutional_layer(l);
+            }
+#endif
+            float zero = 0;
+            fwrite(l.biases, sizeof(float), l.n, fp);
+            fwrite(l.biases, sizeof(float), l.n, fp);
+
+            for (j = 0; j < l.n; ++j){
+                int index = j*l.c*l.size*l.size;
+                fwrite(l.filters+index, sizeof(float), l.c*l.size*l.size, fp);
+                for (k = 0; k < l.c*l.size*l.size; ++k) fwrite(&zero, sizeof(float), 1, fp);
+            }
+            for (j = 0; j < l.n; ++j){
+                int index = j*l.c*l.size*l.size;
+                for (k = 0; k < l.c*l.size*l.size; ++k) fwrite(&zero, sizeof(float), 1, fp);
+                fwrite(l.filters+index, sizeof(float), l.c*l.size*l.size, fp);
+            }
+        }
+    }
+    fclose(fp);
+}
+
+void save_weights_upto(network net, char *filename, int cutoff)
 {
     fprintf(stderr, "Saving weights to %s\n", filename);
     FILE *fp = fopen(filename, "w");
@@ -474,10 +599,10 @@
     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);
+    fwrite(net.seen, sizeof(int), 1, fp);
 
     int i;
-    for(i = 0; i < net.n; ++i){
+    for(i = 0; i < net.n && i < cutoff; ++i){
         layer l = net.layers[i];
         if(l.type == CONVOLUTIONAL){
 #ifdef GPU
@@ -487,19 +612,13 @@
 #endif
             int num = l.n*l.c*l.size*l.size;
             fwrite(l.biases, sizeof(float), l.n, fp);
-            fwrite(l.filters, sizeof(float), num, fp);
-        }
-        if(l.type == DECONVOLUTIONAL){
-#ifdef GPU
-            if(gpu_index >= 0){
-                pull_deconvolutional_layer(l);
+            if (l.batch_normalize){
+                fwrite(l.scales, sizeof(float), l.n, fp);
+                fwrite(l.rolling_mean, sizeof(float), l.n, fp);
+                fwrite(l.rolling_variance, sizeof(float), l.n, fp);
             }
-#endif
-            int num = l.n*l.c*l.size*l.size;
-            fwrite(l.biases, sizeof(float), l.n, fp);
             fwrite(l.filters, sizeof(float), num, fp);
-        }
-        if(l.type == CONNECTED){
+        } if(l.type == CONNECTED){
 #ifdef GPU
             if(gpu_index >= 0){
                 pull_connected_layer(l);
@@ -511,6 +630,10 @@
     }
     fclose(fp);
 }
+void save_weights(network net, char *filename)
+{
+    save_weights_upto(net, filename, net.n);
+}
 
 void load_weights_upto(network *net, char *filename, int cutoff)
 {
@@ -519,17 +642,24 @@
     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);
+    float garbage;
+    fread(&garbage, sizeof(float), 1, fp);
+    fread(&garbage, sizeof(float), 1, fp);
+    fread(&garbage, sizeof(float), 1, fp);
+    fread(net->seen, sizeof(int), 1, fp);
 
     int i;
     for(i = 0; i < net->n && i < cutoff; ++i){
         layer l = net->layers[i];
+        if (l.dontload) continue;
         if(l.type == CONVOLUTIONAL){
             int num = l.n*l.c*l.size*l.size;
             fread(l.biases, sizeof(float), l.n, fp);
+            if (l.batch_normalize && (!l.dontloadscales)){
+                fread(l.scales, sizeof(float), l.n, fp);
+                fread(l.rolling_mean, sizeof(float), l.n, fp);
+                fread(l.rolling_variance, sizeof(float), l.n, fp);
+            }
             fread(l.filters, sizeof(float), num, fp);
 #ifdef GPU
             if(gpu_index >= 0){

--
Gitblit v1.10.0