From 956cfcaec993111426d91bcd61676b5fe0ebfd16 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 24 Feb 2014 21:02:53 +0000
Subject: [PATCH] Feature extraction using Imagenet

---
 src/network.c |  278 ++++++++++++++++++++++++++++++++++++++++++++++---------
 1 files changed, 231 insertions(+), 47 deletions(-)

diff --git a/src/network.c b/src/network.c
index cce673c..b2fc922 100644
--- a/src/network.c
+++ b/src/network.c
@@ -6,6 +6,7 @@
 
 #include "connected_layer.h"
 #include "convolutional_layer.h"
+//#include "old_conv.h"
 #include "maxpool_layer.h"
 #include "softmax_layer.h"
 
@@ -15,10 +16,82 @@
     net.n = n;
     net.layers = calloc(net.n, sizeof(void *));
     net.types = calloc(net.n, sizeof(LAYER_TYPE));
+    net.outputs = 0;
+    net.output = 0;
     return net;
 }
 
-void forward_network(network net, double *input)
+void print_convolutional_cfg(FILE *fp, convolutional_layer *l, int first)
+{
+    int i;
+    fprintf(fp, "[convolutional]\n");
+    if(first) fprintf(fp,   "height=%d\n"
+                            "width=%d\n"
+                            "channels=%d\n",
+                            l->h, l->w, l->c);
+    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, "data=");
+    for(i = 0; i < l->n; ++i) fprintf(fp, "%g,", l->biases[i]);
+    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_connected_cfg(FILE *fp, connected_layer *l, int first)
+{
+    int i;
+    fprintf(fp, "[connected]\n");
+    if(first) fprintf(fp, "input=%d\n", l->inputs);
+    fprintf(fp, "output=%d\n"
+                "activation=%s\n",
+                l->outputs,
+                get_activation_string(l->activation));
+    fprintf(fp, "data=");
+    for(i = 0; i < l->outputs; ++i) fprintf(fp, "%g,", l->biases[i]);
+    for(i = 0; i < l->inputs*l->outputs; ++i) fprintf(fp, "%g,", l->weights[i]);
+    fprintf(fp, "\n\n");
+}
+
+void print_maxpool_cfg(FILE *fp, maxpool_layer *l, int first)
+{
+    fprintf(fp, "[maxpool]\n");
+    if(first) fprintf(fp,   "height=%d\n"
+                            "width=%d\n"
+                            "channels=%d\n",
+                            l->h, l->w, l->c);
+    fprintf(fp, "stride=%d\n\n", l->stride);
+}
+
+void print_softmax_cfg(FILE *fp, softmax_layer *l, int first)
+{
+    fprintf(fp, "[softmax]\n");
+    if(first) fprintf(fp, "input=%d\n", l->inputs);
+    fprintf(fp, "\n");
+}
+
+void save_network(network net, char *filename)
+{
+    FILE *fp = fopen(filename, "w");
+    if(!fp) file_error(filename);
+    int i;
+    for(i = 0; i < net.n; ++i)
+    {
+        if(net.types[i] == CONVOLUTIONAL)
+            print_convolutional_cfg(fp, (convolutional_layer *)net.layers[i], i==0);
+        else if(net.types[i] == CONNECTED)
+            print_connected_cfg(fp, (connected_layer *)net.layers[i], i==0);
+        else if(net.types[i] == MAXPOOL)
+            print_maxpool_cfg(fp, (maxpool_layer *)net.layers[i], i==0);
+        else if(net.types[i] == SOFTMAX)
+            print_softmax_cfg(fp, (softmax_layer *)net.layers[i], i==0);
+    }
+    fclose(fp);
+}
+
+void forward_network(network net, float *input)
 {
     int i;
     for(i = 0; i < net.n; ++i){
@@ -45,13 +118,13 @@
     }
 }
 
-void update_network(network net, double step)
+void update_network(network net, float step, float momentum, float decay)
 {
     int i;
     for(i = 0; i < net.n; ++i){
         if(net.types[i] == CONVOLUTIONAL){
             convolutional_layer layer = *(convolutional_layer *)net.layers[i];
-            update_convolutional_layer(layer, step, 0.9, .01);
+            update_convolutional_layer(layer, step, momentum, decay);
         }
         else if(net.types[i] == MAXPOOL){
             //maxpool_layer layer = *(maxpool_layer *)net.layers[i];
@@ -61,12 +134,12 @@
         }
         else if(net.types[i] == CONNECTED){
             connected_layer layer = *(connected_layer *)net.layers[i];
-            update_connected_layer(layer, step, .9, 0);
+            update_connected_layer(layer, step, momentum, decay);
         }
     }
 }
 
-double *get_network_output_layer(network net, int i)
+float *get_network_output_layer(network net, int i)
 {
     if(net.types[i] == CONVOLUTIONAL){
         convolutional_layer layer = *(convolutional_layer *)net.layers[i];
@@ -83,12 +156,12 @@
     }
     return 0;
 }
-double *get_network_output(network net)
+float *get_network_output(network net)
 {
     return get_network_output_layer(net, net.n-1);
 }
 
-double *get_network_delta_layer(network net, int i)
+float *get_network_delta_layer(network net, int i)
 {
     if(net.types[i] == CONVOLUTIONAL){
         convolutional_layer layer = *(convolutional_layer *)net.layers[i];
@@ -106,16 +179,39 @@
     return 0;
 }
 
-double *get_network_delta(network net)
+float *get_network_delta(network net)
 {
     return get_network_delta_layer(net, net.n-1);
 }
 
-void learn_network(network net, double *input)
+float calculate_error_network(network net, float *truth)
 {
+    float sum = 0;
+    float *delta = get_network_delta(net);
+    float *out = get_network_output(net);
+    int i, k = get_network_output_size(net);
+    for(i = 0; i < k; ++i){
+        printf("%f, ", out[i]);
+        delta[i] = truth[i] - out[i];
+        sum += delta[i]*delta[i];
+    }
+    printf("\n");
+    return sum;
+}
+
+int get_predicted_class_network(network net)
+{
+    float *out = get_network_output(net);
+    int k = get_network_output_size(net);
+    return max_index(out, k);
+}
+
+float backward_network(network net, float *input, float *truth)
+{
+    float error = calculate_error_network(net, truth);
     int i;
-    double *prev_input;
-    double *prev_delta;
+    float *prev_input;
+    float *prev_delta;
     for(i = net.n-1; i >= 0; --i){
         if(i == 0){
             prev_input = input;
@@ -126,8 +222,9 @@
         }
         if(net.types[i] == CONVOLUTIONAL){
             convolutional_layer layer = *(convolutional_layer *)net.layers[i];
-            learn_convolutional_layer(layer, prev_input);
-            if(i != 0) backward_convolutional_layer(layer, prev_input, prev_delta);
+            learn_convolutional_layer(layer);
+            //learn_convolutional_layer(layer);
+            if(i != 0) backward_convolutional_layer(layer, prev_delta);
         }
         else if(net.types[i] == MAXPOOL){
             maxpool_layer layer = *(maxpool_layer *)net.layers[i];
@@ -143,42 +240,71 @@
             if(i != 0) backward_connected_layer(layer, prev_input, prev_delta);
         }
     }
+    return error;
 }
 
-void train_network_batch(network net, batch b)
+float train_network_datum(network net, float *x, float *y, float step, float momentum, float decay)
 {
-    int i,j;
-    int k = get_network_output_size(net);
+    forward_network(net, x);
+    //int class = get_predicted_class_network(net);
+    float error = backward_network(net, x, y);
+    update_network(net, step, momentum, decay);
+    //return (y[class]?1:0);
+    return error;
+}
+
+float train_network_sgd(network net, data d, int n, float step, float momentum,float decay)
+{
+    int i;
+    float error = 0;
     int correct = 0;
-    for(i = 0; i < b.n; ++i){
-        show_image(b.images[i], "Input");
-        forward_network(net, b.images[i].data);
-        image o = get_network_image(net);
-        if(o.h) show_image_collapsed(o, "Output");
-        double *output = get_network_output(net);
-        double *delta = get_network_delta(net);
-        int max_k = 0;
-        double max = 0;
-        for(j = 0; j < k; ++j){
-            delta[j] = b.truth[i][j]-output[j];
-            if(output[j] > max) {
-                max = output[j];
-                max_k = j;
-            }
-        }
-        if(b.truth[i][max_k]) ++correct;
-        printf("%f\n", (double)correct/(i+1));
-        learn_network(net, b.images[i].data);
-        update_network(net, .001);
+    for(i = 0; i < n; ++i){
+        int index = rand()%d.X.rows;
+        error += train_network_datum(net, d.X.vals[index], d.y.vals[index], step, momentum, decay);
+        float *y = d.y.vals[index];
+        int class = get_predicted_class_network(net);
+        correct += (y[class]?1:0);
+        //printf("%d %f %f\n", i,net.output[0], d.y.vals[index][0]);
+        //if((i+1)%10 == 0){
+        //    printf("%d: %f\n", (i+1), (float)correct/(i+1));
+        //}
+    }
+    printf("Accuracy: %f\n",(float) correct/n);
+    return error/n;
+}
+float train_network_batch(network net, data d, int n, float step, float momentum,float decay)
+{
+    int i;
+    int correct = 0;
+    for(i = 0; i < n; ++i){
+        int index = rand()%d.X.rows;
+        float *x = d.X.vals[index];
+        float *y = d.y.vals[index];
+        forward_network(net, x);
+        int class = get_predicted_class_network(net);
+        backward_network(net, x, y);
+        correct += (y[class]?1:0);
+    }
+    update_network(net, step, momentum, decay);
+    return (float)correct/n;
+
+}
+
+
+void train_network(network net, data d, float step, float momentum, float decay)
+{
+    int i;
+    int correct = 0;
+    for(i = 0; i < d.X.rows; ++i){
+        correct += train_network_datum(net, d.X.vals[i], d.y.vals[i], step, momentum, decay);
         if(i%100 == 0){
             visualize_network(net);
-            cvWaitKey(100);
+            cvWaitKey(10);
         }
     }
     visualize_network(net);
-    print_network(net);
     cvWaitKey(100);
-    printf("Accuracy: %f\n", (double)correct/b.n);
+    printf("Accuracy: %f\n", (float)correct/d.X.rows);
 }
 
 int get_network_output_size_layer(network net, int i)
@@ -204,6 +330,34 @@
     return 0;
 }
 
+int reset_network_size(network net, int h, int w, int c)
+{
+    int i;
+    for (i = 0; i < net.n; ++i){
+        if(net.types[i] == CONVOLUTIONAL){
+            convolutional_layer *layer = (convolutional_layer *)net.layers[i];
+            layer->h = h;
+            layer->w = w;
+            layer->c = c;
+            image output = get_convolutional_image(*layer);
+            h = output.h;
+            w = output.w;
+            c = output.c;
+        }
+        else if(net.types[i] == MAXPOOL){
+            maxpool_layer *layer = (maxpool_layer *)net.layers[i];
+            layer->h = h;
+            layer->w = w;
+            layer->c = c;
+            image output = get_maxpool_image(*layer);
+            h = output.h;
+            w = output.w;
+            c = output.c;
+        }
+    }
+    return 0;
+}
+
 int get_network_output_size(network net)
 {
     int i = net.n-1;
@@ -241,16 +395,37 @@
         sprintf(buff, "Layer %d", i);
         if(net.types[i] == CONVOLUTIONAL){
             convolutional_layer layer = *(convolutional_layer *)net.layers[i];
-            visualize_convolutional_filters(layer, buff);
+            visualize_convolutional_layer(layer, buff);
         }
     } 
 }
 
+float *network_predict(network net, float *input)
+{
+    forward_network(net, input);
+    float *out = get_network_output(net);
+    return out;
+}
+
+matrix network_predict_data(network net, data test)
+{
+    int i,j;
+    int k = get_network_output_size(net);
+    matrix pred = make_matrix(test.X.rows, k);
+    for(i = 0; i < test.X.rows; ++i){
+        float *out = network_predict(net, test.X.vals[i]);
+        for(j = 0; j < k; ++j){
+            pred.vals[i][j] = out[j];
+        }
+    }
+    return pred;   
+}
+
 void print_network(network net)
 {
     int i,j;
     for(i = 0; i < net.n; ++i){
-        double *output;
+        float *output = 0;
         int n = 0;
         if(net.types[i] == CONVOLUTIONAL){
             convolutional_layer layer = *(convolutional_layer *)net.layers[i];
@@ -274,12 +449,21 @@
             output = layer.output;
             n = layer.inputs;
         }
-        double mean = mean_array(output, n);
-        double vari = variance_array(output, n);
-        printf("Layer %d - Mean: %f, Variance: %f\n",i,mean, vari);
+        float mean = mean_array(output, n);
+        float vari = variance_array(output, n);
+        fprintf(stderr, "Layer %d - Mean: %f, Variance: %f\n",i,mean, vari);
         if(n > 100) n = 100;
-        for(j = 0; j < n; ++j) printf("%f, ", output[j]);
-        if(n == 100)printf(".....\n");
-        printf("\n");
+        for(j = 0; j < n; ++j) fprintf(stderr, "%f, ", output[j]);
+        if(n == 100)fprintf(stderr,".....\n");
+        fprintf(stderr, "\n");
     }
 }
+
+float network_accuracy(network net, data d)
+{
+    matrix guess = network_predict_data(net, d);
+    float acc = matrix_accuracy(d.y, guess);
+    free_matrix(guess);
+    return acc;
+}
+

--
Gitblit v1.10.0