From f047cfff99e00e28c02eb59b6d32386c122f9af6 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Sun, 08 Mar 2015 18:31:12 +0000
Subject: [PATCH] renamed sigmoid to logistic

---
 src/network.c |  232 +++++++++++++++++++++++++++++++++++++++++++--------------
 1 files changed, 175 insertions(+), 57 deletions(-)

diff --git a/src/network.c b/src/network.c
index f451fd9..b60f059 100644
--- a/src/network.c
+++ b/src/network.c
@@ -8,6 +8,8 @@
 #include "crop_layer.h"
 #include "connected_layer.h"
 #include "convolutional_layer.h"
+#include "deconvolutional_layer.h"
+#include "detection_layer.h"
 #include "maxpool_layer.h"
 #include "cost_layer.h"
 #include "normalization_layer.h"
@@ -15,6 +17,37 @@
 #include "softmax_layer.h"
 #include "dropout_layer.h"
 
+char *get_layer_string(LAYER_TYPE a)
+{
+    switch(a){
+        case CONVOLUTIONAL:
+            return "convolutional";
+        case DECONVOLUTIONAL:
+            return "deconvolutional";
+        case CONNECTED:
+            return "connected";
+        case MAXPOOL:
+            return "maxpool";
+        case SOFTMAX:
+            return "softmax";
+        case DETECTION:
+            return "detection";
+        case NORMALIZATION:
+            return "normalization";
+        case DROPOUT:
+            return "dropout";
+        case FREEWEIGHT:
+            return "freeweight";
+        case CROP:
+            return "crop";
+        case COST:
+            return "cost";
+        default:
+            break;
+    }
+    return "none";
+}
+
 network make_network(int n, int batch)
 {
     network net;
@@ -24,14 +57,14 @@
     net.types = calloc(net.n, sizeof(LAYER_TYPE));
     net.outputs = 0;
     net.output = 0;
+    net.seen = 0;
     #ifdef GPU
-    net.input_cl = calloc(1, sizeof(cl_mem));
-    net.truth_cl = calloc(1, sizeof(cl_mem));
+    net.input_gpu = calloc(1, sizeof(float *));
+    net.truth_gpu = calloc(1, sizeof(float *));
     #endif
     return net;
 }
 
-
 void forward_network(network net, float *input, float *truth, int train)
 {
     int i;
@@ -41,6 +74,16 @@
             forward_convolutional_layer(layer, input);
             input = layer.output;
         }
+        else if(net.types[i] == DECONVOLUTIONAL){
+            deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+            forward_deconvolutional_layer(layer, input);
+            input = layer.output;
+        }
+        else if(net.types[i] == DETECTION){
+            detection_layer layer = *(detection_layer *)net.layers[i];
+            forward_detection_layer(layer, input, truth);
+            input = layer.output;
+        }
         else if(net.types[i] == CONNECTED){
             connected_layer layer = *(connected_layer *)net.layers[i];
             forward_connected_layer(layer, input);
@@ -48,7 +91,7 @@
         }
         else if(net.types[i] == CROP){
             crop_layer layer = *(crop_layer *)net.layers[i];
-            forward_crop_layer(layer, input);
+            forward_crop_layer(layer, train, input);
             input = layer.output;
         }
         else if(net.types[i] == COST){
@@ -74,12 +117,16 @@
             if(!train) continue;
             dropout_layer layer = *(dropout_layer *)net.layers[i];
             forward_dropout_layer(layer, input);
+            input = layer.output;
         }
         else if(net.types[i] == FREEWEIGHT){
             if(!train) continue;
-            freeweight_layer layer = *(freeweight_layer *)net.layers[i];
-            forward_freeweight_layer(layer, input);
+            //freeweight_layer layer = *(freeweight_layer *)net.layers[i];
+            //forward_freeweight_layer(layer, input);
         }
+        //char buff[256];
+        //sprintf(buff, "layer %d", i);
+        //cuda_compare(get_network_output_gpu_layer(net, i), input, get_network_output_size_layer(net, i)*net.batch, buff);
     }
 }
 
@@ -91,14 +138,9 @@
             convolutional_layer layer = *(convolutional_layer *)net.layers[i];
             update_convolutional_layer(layer);
         }
-        else if(net.types[i] == MAXPOOL){
-            //maxpool_layer layer = *(maxpool_layer *)net.layers[i];
-        }
-        else if(net.types[i] == SOFTMAX){
-            //maxpool_layer layer = *(maxpool_layer *)net.layers[i];
-        }
-        else if(net.types[i] == NORMALIZATION){
-            //maxpool_layer layer = *(maxpool_layer *)net.layers[i];
+        else if(net.types[i] == DECONVOLUTIONAL){
+            deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+            update_deconvolutional_layer(layer);
         }
         else if(net.types[i] == CONNECTED){
             connected_layer layer = *(connected_layer *)net.layers[i];
@@ -112,14 +154,21 @@
     if(net.types[i] == CONVOLUTIONAL){
         convolutional_layer layer = *(convolutional_layer *)net.layers[i];
         return layer.output;
+    } else if(net.types[i] == DECONVOLUTIONAL){
+        deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+        return layer.output;
     } else if(net.types[i] == MAXPOOL){
         maxpool_layer layer = *(maxpool_layer *)net.layers[i];
         return layer.output;
+    } else if(net.types[i] == DETECTION){
+        detection_layer layer = *(detection_layer *)net.layers[i];
+        return layer.output;
     } else if(net.types[i] == SOFTMAX){
         softmax_layer layer = *(softmax_layer *)net.layers[i];
         return layer.output;
     } else if(net.types[i] == DROPOUT){
-        return get_network_output_layer(net, i-1);
+        dropout_layer layer = *(dropout_layer *)net.layers[i];
+        return layer.output;
     } else if(net.types[i] == FREEWEIGHT){
         return get_network_output_layer(net, i-1);
     } else if(net.types[i] == CONNECTED){
@@ -146,13 +195,20 @@
     if(net.types[i] == CONVOLUTIONAL){
         convolutional_layer layer = *(convolutional_layer *)net.layers[i];
         return layer.delta;
+    } else if(net.types[i] == DECONVOLUTIONAL){
+        deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+        return layer.delta;
     } else if(net.types[i] == MAXPOOL){
         maxpool_layer layer = *(maxpool_layer *)net.layers[i];
         return layer.delta;
     } else if(net.types[i] == SOFTMAX){
         softmax_layer layer = *(softmax_layer *)net.layers[i];
         return layer.delta;
+    } else if(net.types[i] == DETECTION){
+        detection_layer layer = *(detection_layer *)net.layers[i];
+        return layer.delta;
     } else if(net.types[i] == DROPOUT){
+        if(i == 0) return 0;
         return get_network_delta_layer(net, i-1);
     } else if(net.types[i] == FREEWEIGHT){
         return get_network_delta_layer(net, i-1);
@@ -201,7 +257,7 @@
     return max_index(out, k);
 }
 
-void backward_network(network net, float *input)
+void backward_network(network net, float *input, float *truth)
 {
     int i;
     float *prev_input;
@@ -214,9 +270,13 @@
             prev_input = get_network_output_layer(net, i-1);
             prev_delta = get_network_delta_layer(net, i-1);
         }
+
         if(net.types[i] == CONVOLUTIONAL){
             convolutional_layer layer = *(convolutional_layer *)net.layers[i];
             backward_convolutional_layer(layer, prev_input, prev_delta);
+        } else if(net.types[i] == DECONVOLUTIONAL){
+            deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+            backward_deconvolutional_layer(layer, prev_input, prev_delta);
         }
         else if(net.types[i] == MAXPOOL){
             maxpool_layer layer = *(maxpool_layer *)net.layers[i];
@@ -226,6 +286,10 @@
             dropout_layer layer = *(dropout_layer *)net.layers[i];
             backward_dropout_layer(layer, prev_delta);
         }
+        else if(net.types[i] == DETECTION){
+            detection_layer layer = *(detection_layer *)net.layers[i];
+            backward_detection_layer(layer, prev_input, prev_delta);
+        }
         else if(net.types[i] == NORMALIZATION){
             normalization_layer layer = *(normalization_layer *)net.layers[i];
             if(i != 0) backward_normalization_layer(layer, prev_input, prev_delta);
@@ -245,17 +309,15 @@
     }
 }
 
-
-
-
 float train_network_datum(network net, float *x, float *y)
 {
+    #ifdef GPU
+    if(gpu_index >= 0) return train_network_datum_gpu(net, x, y);
+    #endif
     forward_network(net, x, y, 1);
-    //int class = get_predicted_class_network(net);
-    backward_network(net, x);
+    backward_network(net, x, y);
     float error = get_network_cost(net);
     update_network(net);
-    //return (y[class]?1:0);
     return error;
 }
 
@@ -277,28 +339,10 @@
     return (float)sum/(n*batch);
 }
 
-float train_network_batch(network net, data d, int n)
-{
-    int i,j;
-    float sum = 0;
-    int batch = 2;
-    for(i = 0; i < n; ++i){
-        for(j = 0; j < batch; ++j){
-            int index = rand()%d.X.rows;
-            float *x = d.X.vals[index];
-            float *y = d.y.vals[index];
-            forward_network(net, x, y, 1);
-            backward_network(net, x);
-            sum += get_network_cost(net);
-        }
-        update_network(net);
-    }
-    return (float)sum/(n*batch);
-}
-
-float train_network_data_cpu(network net, data d, int n)
+float train_network(network net, data d)
 {
     int batch = net.batch;
+    int n = d.X.rows / batch;
     float *X = calloc(batch*d.X.cols, sizeof(float));
     float *y = calloc(batch*d.y.cols, sizeof(float));
 
@@ -314,20 +358,23 @@
     return (float)sum/(n*batch);
 }
 
-void train_network(network net, data d)
+float train_network_batch(network net, data d, int n)
 {
-    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]);
-        if(i%100 == 0){
-            visualize_network(net);
-            cvWaitKey(10);
+    int i,j;
+    float sum = 0;
+    int batch = 2;
+    for(i = 0; i < n; ++i){
+        for(j = 0; j < batch; ++j){
+            int index = rand()%d.X.rows;
+            float *x = d.X.vals[index];
+            float *y = d.y.vals[index];
+            forward_network(net, x, y, 1);
+            backward_network(net, x, y);
+            sum += get_network_cost(net);
         }
+        update_network(net);
     }
-    visualize_network(net);
-    cvWaitKey(100);
-    fprintf(stderr, "Accuracy: %f\n", (float)correct/d.X.rows);
+    return (float)sum/(n*batch);
 }
 
 void set_learning_network(network *net, float rate, float momentum, float decay)
@@ -352,7 +399,6 @@
     }
 }
 
-
 void set_batch_network(network *net, int b)
 {
     net->batch = b;
@@ -361,6 +407,9 @@
         if(net->types[i] == CONVOLUTIONAL){
             convolutional_layer *layer = (convolutional_layer *)net->layers[i];
             layer->batch = b;
+        }else if(net->types[i] == DECONVOLUTIONAL){
+            deconvolutional_layer *layer = (deconvolutional_layer *)net->layers[i];
+            layer->batch = b;
         }
         else if(net->types[i] == MAXPOOL){
             maxpool_layer *layer = (maxpool_layer *)net->layers[i];
@@ -372,6 +421,9 @@
         } else if(net->types[i] == DROPOUT){
             dropout_layer *layer = (dropout_layer *) net->layers[i];
             layer->batch = b;
+        } else if(net->types[i] == DETECTION){
+            detection_layer *layer = (detection_layer *) net->layers[i];
+            layer->batch = b;
         }
         else if(net->types[i] == FREEWEIGHT){
             freeweight_layer *layer = (freeweight_layer *) net->layers[i];
@@ -385,6 +437,10 @@
             cost_layer *layer = (cost_layer *)net->layers[i];
             layer->batch = b;
         }
+        else if(net->types[i] == CROP){
+            crop_layer *layer = (crop_layer *)net->layers[i];
+            layer->batch = b;
+        }
     }
 }
 
@@ -395,6 +451,10 @@
         convolutional_layer layer = *(convolutional_layer *)net.layers[i];
         return layer.h*layer.w*layer.c;
     }
+    if(net.types[i] == DECONVOLUTIONAL){
+        deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+        return layer.h*layer.w*layer.c;
+    }
     else if(net.types[i] == MAXPOOL){
         maxpool_layer layer = *(maxpool_layer *)net.layers[i];
         return layer.h*layer.w*layer.c;
@@ -405,6 +465,9 @@
     } else if(net.types[i] == DROPOUT){
         dropout_layer layer = *(dropout_layer *) net.layers[i];
         return layer.inputs;
+    } else if(net.types[i] == DETECTION){
+        detection_layer layer = *(detection_layer *) net.layers[i];
+        return layer.inputs;
     } else if(net.types[i] == CROP){
         crop_layer layer = *(crop_layer *) net.layers[i];
         return layer.c*layer.h*layer.w;
@@ -428,6 +491,15 @@
         image output = get_convolutional_image(layer);
         return output.h*output.w*output.c;
     }
+    else if(net.types[i] == DECONVOLUTIONAL){
+        deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+        image output = get_deconvolutional_image(layer);
+        return output.h*output.w*output.c;
+    }
+    else if(net.types[i] == DETECTION){
+        detection_layer layer = *(detection_layer *)net.layers[i];
+        return get_detection_layer_output_size(layer);
+    }
     else if(net.types[i] == MAXPOOL){
         maxpool_layer layer = *(maxpool_layer *)net.layers[i];
         image output = get_maxpool_image(layer);
@@ -463,21 +535,31 @@
     for (i = 0; i < net.n; ++i){
         if(net.types[i] == CONVOLUTIONAL){
             convolutional_layer *layer = (convolutional_layer *)net.layers[i];
-            resize_convolutional_layer(layer, h, w, c);
+            resize_convolutional_layer(layer, h, w);
             image output = get_convolutional_image(*layer);
             h = output.h;
             w = output.w;
             c = output.c;
+        } else if(net.types[i] == DECONVOLUTIONAL){
+            deconvolutional_layer *layer = (deconvolutional_layer *)net.layers[i];
+            resize_deconvolutional_layer(layer, h, w);
+            image output = get_deconvolutional_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];
-            resize_maxpool_layer(layer, h, w, c);
+            resize_maxpool_layer(layer, h, w);
             image output = get_maxpool_image(*layer);
             h = output.h;
             w = output.w;
             c = output.c;
+        }else if(net.types[i] == DROPOUT){
+            dropout_layer *layer = (dropout_layer *)net.layers[i];
+            resize_dropout_layer(layer, h*w*c);
         }else if(net.types[i] == NORMALIZATION){
             normalization_layer *layer = (normalization_layer *)net.layers[i];
-            resize_normalization_layer(layer, h, w, c);
+            resize_normalization_layer(layer, h, w);
             image output = get_normalization_image(*layer);
             h = output.h;
             w = output.w;
@@ -507,6 +589,10 @@
         convolutional_layer layer = *(convolutional_layer *)net.layers[i];
         return get_convolutional_image(layer);
     }
+    else if(net.types[i] == DECONVOLUTIONAL){
+        deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
+        return get_deconvolutional_image(layer);
+    }
     else if(net.types[i] == MAXPOOL){
         maxpool_layer layer = *(maxpool_layer *)net.layers[i];
         return get_maxpool_image(layer);
@@ -515,6 +601,9 @@
         normalization_layer layer = *(normalization_layer *)net.layers[i];
         return get_normalization_image(layer);
     }
+    else if(net.types[i] == DROPOUT){
+        return get_network_image_layer(net, i-1);
+    }
     else if(net.types[i] == CROP){
         crop_layer layer = *(crop_layer *)net.layers[i];
         return get_crop_image(layer);
@@ -561,6 +650,10 @@
 
 float *network_predict(network net, float *input)
 {
+    #ifdef GPU
+    if(gpu_index >= 0)  return network_predict_gpu(net, input);
+    #endif
+
     forward_network(net, input, 0, 0);
     float *out = get_network_output(net);
     return out;
@@ -658,6 +751,31 @@
     }
 }
 
+void compare_networks(network n1, network n2, data test)
+{
+    matrix g1 = network_predict_data(n1, test);
+    matrix g2 = network_predict_data(n2, test);
+    int i;
+    int a,b,c,d;
+    a = b = c = d = 0;
+    for(i = 0; i < g1.rows; ++i){
+        int truth = max_index(test.y.vals[i], test.y.cols);
+        int p1 = max_index(g1.vals[i], g1.cols);
+        int p2 = max_index(g2.vals[i], g2.cols);
+        if(p1 == truth){
+            if(p2 == truth) ++d;
+            else ++c;
+        }else{
+            if(p2 == truth) ++b;
+            else ++a;
+        }
+    }
+    printf("%5d %5d\n%5d %5d\n", a, b, c, d);
+    float num = pow((abs(b - c) - 1.), 2.);
+    float den = b + c;
+    printf("%f\n", num/den); 
+}
+
 float network_accuracy(network net, data d)
 {
     matrix guess = network_predict_data(net, d);

--
Gitblit v1.10.0