From 1edcf73a73d2007afc61289245763f5cf0c29e10 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Thu, 04 Dec 2014 07:20:29 +0000
Subject: [PATCH] Detection good, split up col images

---
 src/network.c |  232 ++++++++++++++++++++++++++++++++++-----------------------
 1 files changed, 139 insertions(+), 93 deletions(-)

diff --git a/src/network.c b/src/network.c
index ed927a8..3a6a184 100644
--- a/src/network.c
+++ b/src/network.c
@@ -1,13 +1,17 @@
 #include <stdio.h>
+#include <time.h>
 #include "network.h"
 #include "image.h"
 #include "data.h"
 #include "utils.h"
 
+#include "crop_layer.h"
 #include "connected_layer.h"
 #include "convolutional_layer.h"
 #include "maxpool_layer.h"
+#include "cost_layer.h"
 #include "normalization_layer.h"
+#include "freeweight_layer.h"
 #include "softmax_layer.h"
 #include "dropout_layer.h"
 
@@ -21,57 +25,14 @@
     net.outputs = 0;
     net.output = 0;
     #ifdef GPU
-    net.input_cl = 0;
+    net.input_cl = calloc(1, sizeof(cl_mem));
+    net.truth_cl = calloc(1, sizeof(cl_mem));
     #endif
     return net;
 }
 
-#ifdef GPU
-void forward_network(network net, float *input, int train)
-{
-    cl_setup();
-    size_t size = get_network_input_size(net);
-    if(!net.input_cl){
-        net.input_cl = clCreateBuffer(cl.context,
-            CL_MEM_READ_WRITE, size*sizeof(float), 0, &cl.error);
-        check_error(cl);
-    }
-    cl_write_array(net.input_cl, input, size);
-    cl_mem input_cl = net.input_cl;
-    int i;
-    for(i = 0; i < net.n; ++i){
-        if(net.types[i] == CONVOLUTIONAL){
-            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
-            forward_convolutional_layer_gpu(layer, input_cl);
-            input_cl = layer.output_cl;
-            input = layer.output;
-        }
-        else if(net.types[i] == CONNECTED){
-            connected_layer layer = *(connected_layer *)net.layers[i];
-            forward_connected_layer(layer, input, train);
-            input = layer.output;
-        }
-        else if(net.types[i] == SOFTMAX){
-            softmax_layer layer = *(softmax_layer *)net.layers[i];
-            forward_softmax_layer(layer, input);
-            input = layer.output;
-        }
-        else if(net.types[i] == MAXPOOL){
-            maxpool_layer layer = *(maxpool_layer *)net.layers[i];
-            forward_maxpool_layer(layer, input);
-            input = layer.output;
-        }
-        else if(net.types[i] == NORMALIZATION){
-            normalization_layer layer = *(normalization_layer *)net.layers[i];
-            forward_normalization_layer(layer, input);
-            input = layer.output;
-        }
-    }
-}
 
-#else
-
-void forward_network(network net, float *input, int train)
+void forward_network(network net, float *input, float *truth, int train)
 {
     int i;
     for(i = 0; i < net.n; ++i){
@@ -85,6 +46,15 @@
             forward_connected_layer(layer, input);
             input = layer.output;
         }
+        else if(net.types[i] == CROP){
+            crop_layer layer = *(crop_layer *)net.layers[i];
+            forward_crop_layer(layer, input);
+            input = layer.output;
+        }
+        else if(net.types[i] == COST){
+            cost_layer layer = *(cost_layer *)net.layers[i];
+            forward_cost_layer(layer, input, truth);
+        }
         else if(net.types[i] == SOFTMAX){
             softmax_layer layer = *(softmax_layer *)net.layers[i];
             forward_softmax_layer(layer, input);
@@ -105,9 +75,13 @@
             dropout_layer layer = *(dropout_layer *)net.layers[i];
             forward_dropout_layer(layer, input);
         }
+        else if(net.types[i] == FREEWEIGHT){
+            if(!train) continue;
+            freeweight_layer layer = *(freeweight_layer *)net.layers[i];
+            forward_freeweight_layer(layer, input);
+        }
     }
 }
-#endif
 
 void update_network(network net)
 {
@@ -146,6 +120,8 @@
         return layer.output;
     } else if(net.types[i] == DROPOUT){
         return get_network_output_layer(net, i-1);
+    } else if(net.types[i] == FREEWEIGHT){
+        return get_network_output_layer(net, i-1);
     } else if(net.types[i] == CONNECTED){
         connected_layer layer = *(connected_layer *)net.layers[i];
         return layer.output;
@@ -157,7 +133,9 @@
 }
 float *get_network_output(network net)
 {
-    return get_network_output_layer(net, net.n-1);
+    int i;
+    for(i = net.n-1; i > 0; --i) if(net.types[i] != COST) break;
+    return get_network_output_layer(net, i);
 }
 
 float *get_network_delta_layer(network net, int i)
@@ -173,6 +151,8 @@
         return layer.delta;
     } else if(net.types[i] == DROPOUT){
         return get_network_delta_layer(net, i-1);
+    } else if(net.types[i] == FREEWEIGHT){
+        return get_network_delta_layer(net, i-1);
     } else if(net.types[i] == CONNECTED){
         connected_layer layer = *(connected_layer *)net.layers[i];
         return layer.delta;
@@ -180,6 +160,14 @@
     return 0;
 }
 
+float get_network_cost(network net)
+{
+    if(net.types[net.n-1] == COST){
+        return ((cost_layer *)net.layers[net.n-1])->output[0];
+    }
+    return 0;
+}
+
 float *get_network_delta(network net)
 {
     return get_network_delta_layer(net, net.n-1);
@@ -210,9 +198,8 @@
     return max_index(out, k);
 }
 
-float backward_network(network net, float *input, float *truth)
+void backward_network(network net, float *input)
 {
-    float error = calculate_error_network(net, truth);
     int i;
     float *prev_input;
     float *prev_delta;
@@ -226,11 +213,11 @@
         }
         if(net.types[i] == CONVOLUTIONAL){
             convolutional_layer layer = *(convolutional_layer *)net.layers[i];
-            backward_convolutional_layer(layer, prev_delta);
+            backward_convolutional_layer(layer, prev_input, prev_delta);
         }
         else if(net.types[i] == MAXPOOL){
             maxpool_layer layer = *(maxpool_layer *)net.layers[i];
-            if(i != 0) backward_maxpool_layer(layer, prev_input, prev_delta);
+            if(i != 0) backward_maxpool_layer(layer, prev_delta);
         }
         else if(net.types[i] == NORMALIZATION){
             normalization_layer layer = *(normalization_layer *)net.layers[i];
@@ -238,21 +225,28 @@
         }
         else if(net.types[i] == SOFTMAX){
             softmax_layer layer = *(softmax_layer *)net.layers[i];
-            if(i != 0) backward_softmax_layer(layer, prev_input, prev_delta);
+            if(i != 0) backward_softmax_layer(layer, prev_delta);
         }
         else if(net.types[i] == CONNECTED){
             connected_layer layer = *(connected_layer *)net.layers[i];
             backward_connected_layer(layer, prev_input, prev_delta);
         }
+        else if(net.types[i] == COST){
+            cost_layer layer = *(cost_layer *)net.layers[i];
+            backward_cost_layer(layer, prev_input, prev_delta);
+        }
     }
-    return error;
 }
 
+
+
+
 float train_network_datum(network net, float *x, float *y)
 {
-    forward_network(net, x, 1);
+    forward_network(net, x, y, 1);
     //int class = get_predicted_class_network(net);
-    float error = backward_network(net, x, y);
+    backward_network(net, x);
+    float error = get_network_cost(net);
     update_network(net);
     //return (y[class]?1:0);
     return error;
@@ -264,46 +258,18 @@
     float *X = calloc(batch*d.X.cols, sizeof(float));
     float *y = calloc(batch*d.y.cols, sizeof(float));
 
-    int i,j;
+    int i;
     float sum = 0;
     for(i = 0; i < n; ++i){
-        for(j = 0; j < batch; ++j){
-            int index = rand()%d.X.rows;
-            memcpy(X+j*d.X.cols, d.X.vals[index], d.X.cols*sizeof(float));
-            memcpy(y+j*d.y.cols, d.y.vals[index], d.y.cols*sizeof(float));
-        }
+        get_random_batch(d, batch, X, y);
         float err = train_network_datum(net, X, y);
         sum += err;
-        //train_network_datum(net, X, y);
-        /*
-        float *y = d.y.vals[index];
-        int class = get_predicted_class_network(net);
-        correct += (y[class]?1:0);
-        */
-
-/*
-        for(j = 0; j < d.y.cols*batch; ++j){
-            printf("%6.3f ", y[j]);
-        }
-        printf("\n");
-        for(j = 0; j < d.y.cols*batch; ++j){
-            printf("%6.3f ", get_network_output(net)[j]);
-        }
-        printf("\n");
-        printf("\n");
-        */
-
-
-        //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);
     free(X);
     free(y);
     return (float)sum/(n*batch);
 }
+
 float train_network_batch(network net, data d, int n)
 {
     int i,j;
@@ -314,14 +280,32 @@
             int index = rand()%d.X.rows;
             float *x = d.X.vals[index];
             float *y = d.y.vals[index];
-            forward_network(net, x, 1);
-            sum += backward_network(net, x, y);
+            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)
+{
+    int batch = net.batch;
+    float *X = calloc(batch*d.X.cols, sizeof(float));
+    float *y = calloc(batch*d.y.cols, sizeof(float));
+
+    int i;
+    float sum = 0;
+    for(i = 0; i < n; ++i){
+        get_next_batch(d, batch, i*batch, X, y);
+        float err = train_network_datum(net, X, y);
+        sum += err;
+    }
+    free(X);
+    free(y);
+    return (float)sum/(n*batch);
+}
 
 void train_network(network net, data d)
 {
@@ -356,6 +340,10 @@
         dropout_layer layer = *(dropout_layer *) net.layers[i];
         return layer.inputs;
     }
+    else if(net.types[i] == FREEWEIGHT){
+        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
+        return layer.inputs;
+    }
     else if(net.types[i] == SOFTMAX){
         softmax_layer layer = *(softmax_layer *)net.layers[i];
         return layer.inputs;
@@ -378,10 +366,15 @@
     else if(net.types[i] == CONNECTED){
         connected_layer layer = *(connected_layer *)net.layers[i];
         return layer.outputs;
-    } else if(net.types[i] == DROPOUT){
+    }
+    else if(net.types[i] == DROPOUT){
         dropout_layer layer = *(dropout_layer *) net.layers[i];
         return layer.inputs;
     }
+    else if(net.types[i] == FREEWEIGHT){
+        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
+        return layer.inputs;
+    }
     else if(net.types[i] == SOFTMAX){
         softmax_layer layer = *(softmax_layer *)net.layers[i];
         return layer.inputs;
@@ -423,7 +416,8 @@
 
 int get_network_output_size(network net)
 {
-    int i = net.n-1;
+    int i;
+    for(i = net.n-1; i > 0; --i) if(net.types[i] != COST) break;
     return get_network_output_size_layer(net, i);
 }
 
@@ -446,6 +440,10 @@
         normalization_layer layer = *(normalization_layer *)net.layers[i];
         return get_normalization_image(layer);
     }
+    else if(net.types[i] == CROP){
+        crop_layer layer = *(crop_layer *)net.layers[i];
+        return get_crop_image(layer);
+    }
     return make_empty_image(0,0,0);
 }
 
@@ -464,6 +462,7 @@
     image *prev = 0;
     int i;
     char buff[256];
+    //show_image(get_network_image_layer(net, 0), "Crop");
     for(i = 0; i < net.n; ++i){
         sprintf(buff, "Layer %d", i);
         if(net.types[i] == CONVOLUTIONAL){
@@ -477,19 +476,52 @@
     } 
 }
 
+void top_predictions(network net, int k, int *index)
+{
+    int size = get_network_output_size(net);
+    float *out = get_network_output(net);
+    top_k(out, size, k, index);
+}
+
+
 float *network_predict(network net, float *input)
 {
-    forward_network(net, input, 0);
+    forward_network(net, input, 0, 0);
     float *out = get_network_output(net);
     return out;
 }
 
+matrix network_predict_data_multi(network net, data test, int n)
+{
+    int i,j,b,m;
+    int k = get_network_output_size(net);
+    matrix pred = make_matrix(test.X.rows, k);
+    float *X = calloc(net.batch*test.X.rows, sizeof(float));
+    for(i = 0; i < test.X.rows; i += net.batch){
+        for(b = 0; b < net.batch; ++b){
+            if(i+b == test.X.rows) break;
+            memcpy(X+b*test.X.cols, test.X.vals[i+b], test.X.cols*sizeof(float));
+        }
+        for(m = 0; m < n; ++m){
+            float *out = network_predict(net, X);
+            for(b = 0; b < net.batch; ++b){
+                if(i+b == test.X.rows) break;
+                for(j = 0; j < k; ++j){
+                    pred.vals[i+b][j] += out[j+b*k]/n;
+                }
+            }
+        }
+    }
+    free(X);
+    return pred;   
+}
+
 matrix network_predict_data(network net, data test)
 {
     int i,j,b;
     int k = get_network_output_size(net);
     matrix pred = make_matrix(test.X.rows, k);
-    float *X = calloc(net.batch*test.X.rows, sizeof(float));
+    float *X = calloc(net.batch*test.X.cols, sizeof(float));
     for(i = 0; i < test.X.rows; i += net.batch){
         for(b = 0; b < net.batch; ++b){
             if(i+b == test.X.rows) break;
@@ -525,6 +557,12 @@
             image m = get_maxpool_image(layer);
             n = m.h*m.w*m.c;
         }
+        else if(net.types[i] == CROP){
+            crop_layer layer = *(crop_layer *)net.layers[i];
+            output = layer.output;
+            image m = get_crop_image(layer);
+            n = m.h*m.w*m.c;
+        }
         else if(net.types[i] == CONNECTED){
             connected_layer layer = *(connected_layer *)net.layers[i];
             output = layer.output;
@@ -553,4 +591,12 @@
     return acc;
 }
 
+float network_accuracy_multi(network net, data d, int n)
+{
+    matrix guess = network_predict_data_multi(net, d, n);
+    float acc = matrix_accuracy(d.y, guess);
+    free_matrix(guess);
+    return acc;
+}
+
 

--
Gitblit v1.10.0