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/network.c | 740 ++++++++++++++++++++------------------------------------
1 files changed, 267 insertions(+), 473 deletions(-)
diff --git a/src/network.c b/src/network.c
index b60f059..8dee8cc 100644
--- a/src/network.c
+++ b/src/network.c
@@ -4,60 +4,113 @@
#include "image.h"
#include "data.h"
#include "utils.h"
+#include "blas.h"
#include "crop_layer.h"
#include "connected_layer.h"
+#include "local_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"
-#include "freeweight_layer.h"
+#include "maxpool_layer.h"
+#include "avgpool_layer.h"
+#include "cost_layer.h"
#include "softmax_layer.h"
#include "dropout_layer.h"
+#include "route_layer.h"
+#include "shortcut_layer.h"
+
+int get_current_batch(network net)
+{
+ int batch_num = (*net.seen)/(net.batch*net.subdivisions);
+ return batch_num;
+}
+
+void reset_momentum(network net)
+{
+ if (net.momentum == 0) return;
+ net.learning_rate = 0;
+ net.momentum = 0;
+ net.decay = 0;
+ #ifdef GPU
+ if(gpu_index >= 0) update_network_gpu(net);
+ #endif
+}
+
+float get_current_rate(network net)
+{
+ int batch_num = get_current_batch(net);
+ int i;
+ float rate;
+ switch (net.policy) {
+ case CONSTANT:
+ return net.learning_rate;
+ case STEP:
+ return net.learning_rate * pow(net.scale, batch_num/net.step);
+ case STEPS:
+ rate = net.learning_rate;
+ for(i = 0; i < net.num_steps; ++i){
+ if(net.steps[i] > batch_num) return rate;
+ rate *= net.scales[i];
+ if(net.steps[i] > batch_num - 1) reset_momentum(net);
+ }
+ return rate;
+ case EXP:
+ return net.learning_rate * pow(net.gamma, batch_num);
+ case POLY:
+ return net.learning_rate * pow(1 - (float)batch_num / net.max_batches, net.power);
+ case SIG:
+ return net.learning_rate * (1./(1.+exp(net.gamma*(batch_num - net.step))));
+ default:
+ fprintf(stderr, "Policy is weird!\n");
+ return net.learning_rate;
+ }
+}
char *get_layer_string(LAYER_TYPE a)
{
switch(a){
case CONVOLUTIONAL:
return "convolutional";
+ case LOCAL:
+ return "local";
case DECONVOLUTIONAL:
return "deconvolutional";
case CONNECTED:
return "connected";
case MAXPOOL:
return "maxpool";
+ case AVGPOOL:
+ return "avgpool";
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";
+ case ROUTE:
+ return "route";
+ case SHORTCUT:
+ return "shortcut";
+ case NORMALIZATION:
+ return "normalization";
default:
break;
}
return "none";
}
-network make_network(int n, int batch)
+network make_network(int n)
{
- network net;
+ network net = {0};
net.n = n;
- net.batch = batch;
- net.layers = calloc(net.n, sizeof(void *));
- net.types = calloc(net.n, sizeof(LAYER_TYPE));
- net.outputs = 0;
- net.output = 0;
- net.seen = 0;
+ net.layers = calloc(net.n, sizeof(layer));
+ net.seen = calloc(1, sizeof(int));
#ifdef GPU
net.input_gpu = calloc(1, sizeof(float *));
net.truth_gpu = calloc(1, sizeof(float *));
@@ -65,189 +118,90 @@
return net;
}
-void forward_network(network net, float *input, float *truth, int train)
+void forward_network(network net, network_state state)
{
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(layer, input);
- input = layer.output;
+ state.index = i;
+ layer l = net.layers[i];
+ if(l.delta){
+ scal_cpu(l.outputs * l.batch, 0, l.delta, 1);
}
- else if(net.types[i] == DECONVOLUTIONAL){
- deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
- forward_deconvolutional_layer(layer, input);
- input = layer.output;
+ if(l.type == CONVOLUTIONAL){
+ forward_convolutional_layer(l, state);
+ } else if(l.type == DECONVOLUTIONAL){
+ forward_deconvolutional_layer(l, state);
+ } else if(l.type == LOCAL){
+ forward_local_layer(l, state);
+ } else if(l.type == NORMALIZATION){
+ forward_normalization_layer(l, state);
+ } else if(l.type == DETECTION){
+ forward_detection_layer(l, state);
+ } else if(l.type == CONNECTED){
+ forward_connected_layer(l, state);
+ } else if(l.type == CROP){
+ forward_crop_layer(l, state);
+ } else if(l.type == COST){
+ forward_cost_layer(l, state);
+ } else if(l.type == SOFTMAX){
+ forward_softmax_layer(l, state);
+ } else if(l.type == MAXPOOL){
+ forward_maxpool_layer(l, state);
+ } else if(l.type == AVGPOOL){
+ forward_avgpool_layer(l, state);
+ } else if(l.type == DROPOUT){
+ forward_dropout_layer(l, state);
+ } else if(l.type == ROUTE){
+ forward_route_layer(l, net);
+ } else if(l.type == SHORTCUT){
+ forward_shortcut_layer(l, state);
}
- 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);
- input = layer.output;
- }
- else if(net.types[i] == CROP){
- crop_layer layer = *(crop_layer *)net.layers[i];
- forward_crop_layer(layer, train, 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);
- 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 if(net.types[i] == DROPOUT){
- 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);
- }
- //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);
+ state.input = l.output;
}
}
void update_network(network net)
{
int i;
+ int update_batch = net.batch*net.subdivisions;
+ float rate = get_current_rate(net);
for(i = 0; i < net.n; ++i){
- if(net.types[i] == CONVOLUTIONAL){
- convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- update_convolutional_layer(layer);
- }
- 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];
- update_connected_layer(layer);
+ layer l = net.layers[i];
+ if(l.type == CONVOLUTIONAL){
+ update_convolutional_layer(l, update_batch, rate, net.momentum, net.decay);
+ } else if(l.type == DECONVOLUTIONAL){
+ update_deconvolutional_layer(l, rate, net.momentum, net.decay);
+ } else if(l.type == CONNECTED){
+ update_connected_layer(l, update_batch, rate, net.momentum, net.decay);
+ } else if(l.type == LOCAL){
+ update_local_layer(l, update_batch, rate, net.momentum, net.decay);
}
}
}
-float *get_network_output_layer(network net, int i)
-{
- 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){
- 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){
- connected_layer layer = *(connected_layer *)net.layers[i];
- return layer.output;
- } else if(net.types[i] == CROP){
- crop_layer layer = *(crop_layer *)net.layers[i];
- return layer.output;
- } else if(net.types[i] == NORMALIZATION){
- normalization_layer layer = *(normalization_layer *)net.layers[i];
- return layer.output;
- }
- return 0;
-}
float *get_network_output(network net)
{
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)
-{
- 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);
- } else if(net.types[i] == CONNECTED){
- connected_layer layer = *(connected_layer *)net.layers[i];
- return layer.delta;
- }
- return 0;
+ for(i = net.n-1; i > 0; --i) if(net.layers[i].type != COST) break;
+ return net.layers[i].output;
}
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);
-}
-
-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;
- for(i = 0; i < get_network_output_size(net)*net.batch; ++i){
- //if(i %get_network_output_size(net) == 0) printf("\n");
- //printf("%5.2f %5.2f, ", out[i], truth[i]);
- //if(i == get_network_output_size(net)) printf("\n");
- delta[i] = truth[i] - out[i];
- //printf("%.10f, ", out[i]);
- sum += delta[i]*delta[i];
+ float sum = 0;
+ int count = 0;
+ for(i = 0; i < net.n; ++i){
+ if(net.layers[i].type == COST){
+ sum += net.layers[i].output[0];
+ ++count;
+ }
+ if(net.layers[i].type == DETECTION){
+ sum += net.layers[i].cost[0];
+ ++count;
+ }
}
- //printf("\n");
- return sum;
+ return sum/count;
}
int get_predicted_class_network(network net)
@@ -257,67 +211,69 @@
return max_index(out, k);
}
-void backward_network(network net, float *input, float *truth)
+void backward_network(network net, network_state state)
{
int i;
- float *prev_input;
- float *prev_delta;
+ float *original_input = state.input;
+ float *original_delta = state.delta;
for(i = net.n-1; i >= 0; --i){
+ state.index = i;
if(i == 0){
- prev_input = input;
- prev_delta = 0;
+ state.input = original_input;
+ state.delta = original_delta;
}else{
- prev_input = get_network_output_layer(net, i-1);
- prev_delta = get_network_delta_layer(net, i-1);
+ layer prev = net.layers[i-1];
+ state.input = prev.output;
+ state.delta = prev.delta;
}
-
- 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];
- if(i != 0) backward_maxpool_layer(layer, prev_delta);
- }
- else if(net.types[i] == DROPOUT){
- 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);
- }
- else if(net.types[i] == SOFTMAX){
- softmax_layer layer = *(softmax_layer *)net.layers[i];
- 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);
+ layer l = net.layers[i];
+ if(l.type == CONVOLUTIONAL){
+ backward_convolutional_layer(l, state);
+ } else if(l.type == DECONVOLUTIONAL){
+ backward_deconvolutional_layer(l, state);
+ } else if(l.type == NORMALIZATION){
+ backward_normalization_layer(l, state);
+ } else if(l.type == MAXPOOL){
+ if(i != 0) backward_maxpool_layer(l, state);
+ } else if(l.type == AVGPOOL){
+ backward_avgpool_layer(l, state);
+ } else if(l.type == DROPOUT){
+ backward_dropout_layer(l, state);
+ } else if(l.type == DETECTION){
+ backward_detection_layer(l, state);
+ } else if(l.type == SOFTMAX){
+ if(i != 0) backward_softmax_layer(l, state);
+ } else if(l.type == CONNECTED){
+ backward_connected_layer(l, state);
+ } else if(l.type == LOCAL){
+ backward_local_layer(l, state);
+ } else if(l.type == COST){
+ backward_cost_layer(l, state);
+ } else if(l.type == ROUTE){
+ backward_route_layer(l, net);
+ } else if(l.type == SHORTCUT){
+ backward_shortcut_layer(l, state);
}
}
}
float train_network_datum(network net, float *x, float *y)
{
- #ifdef GPU
+ *net.seen += net.batch;
+#ifdef GPU
if(gpu_index >= 0) return train_network_datum_gpu(net, x, y);
- #endif
- forward_network(net, x, y, 1);
- backward_network(net, x, y);
+#endif
+ network_state state;
+ state.index = 0;
+ state.net = net;
+ state.input = x;
+ state.delta = 0;
+ state.truth = y;
+ state.train = 1;
+ forward_network(net, state);
+ backward_network(net, state);
float error = get_network_cost(net);
- update_network(net);
+ if(((*net.seen)/net.batch)%net.subdivisions == 0) update_network(net);
return error;
}
@@ -361,15 +317,20 @@
float train_network_batch(network net, data d, int n)
{
int i,j;
+ network_state state;
+ state.index = 0;
+ state.net = net;
+ state.train = 1;
+ state.delta = 0;
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);
+ state.input = d.X.vals[index];
+ state.truth = d.y.vals[index];
+ forward_network(net, state);
+ backward_network(net, state);
sum += get_network_cost(net);
}
update_network(net);
@@ -377,238 +338,82 @@
return (float)sum/(n*batch);
}
-void set_learning_network(network *net, float rate, float momentum, float decay)
-{
- int i;
- net->learning_rate=rate;
- net->momentum = momentum;
- net->decay = decay;
- for(i = 0; i < net->n; ++i){
- if(net->types[i] == CONVOLUTIONAL){
- convolutional_layer *layer = (convolutional_layer *)net->layers[i];
- layer->learning_rate=rate;
- layer->momentum = momentum;
- layer->decay = decay;
- }
- else if(net->types[i] == CONNECTED){
- connected_layer *layer = (connected_layer *)net->layers[i];
- layer->learning_rate=rate;
- layer->momentum = momentum;
- layer->decay = decay;
- }
- }
-}
-
void set_batch_network(network *net, int b)
{
net->batch = b;
int i;
for(i = 0; i < net->n; ++i){
- 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];
- layer->batch = b;
- }
- else if(net->types[i] == CONNECTED){
- connected_layer *layer = (connected_layer *)net->layers[i];
- layer->batch = b;
- } 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];
- layer->batch = b;
- }
- else if(net->types[i] == SOFTMAX){
- softmax_layer *layer = (softmax_layer *)net->layers[i];
- layer->batch = b;
- }
- else if(net->types[i] == COST){
- 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;
- }
+ net->layers[i].batch = b;
}
}
-
-int get_network_input_size_layer(network net, int i)
-{
- if(net.types[i] == CONVOLUTIONAL){
- 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;
- }
- else if(net.types[i] == CONNECTED){
- connected_layer layer = *(connected_layer *)net.layers[i];
- return layer.inputs;
- } 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;
- }
- 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;
- }
- printf("Can't find input size\n");
- return 0;
-}
-
-int get_network_output_size_layer(network net, int i)
-{
- if(net.types[i] == CONVOLUTIONAL){
- convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- 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);
- return output.h*output.w*output.c;
- }
- else if(net.types[i] == CROP){
- crop_layer layer = *(crop_layer *) net.layers[i];
- return layer.c*layer.crop_height*layer.crop_width;
- }
- else if(net.types[i] == CONNECTED){
- connected_layer layer = *(connected_layer *)net.layers[i];
- return layer.outputs;
- }
- 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;
- }
- printf("Can't find output size\n");
- return 0;
-}
-
-int resize_network(network net, int h, int w, int c)
+int resize_network(network *net, int w, int h)
{
int i;
- 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);
- 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);
- 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);
- image output = get_normalization_image(*layer);
- h = output.h;
- w = output.w;
- c = output.c;
+ //if(w == net->w && h == net->h) return 0;
+ net->w = w;
+ net->h = h;
+ int inputs = 0;
+ //fprintf(stderr, "Resizing to %d x %d...", w, h);
+ //fflush(stderr);
+ for (i = 0; i < net->n; ++i){
+ layer l = net->layers[i];
+ if(l.type == CONVOLUTIONAL){
+ resize_convolutional_layer(&l, w, h);
+ }else if(l.type == MAXPOOL){
+ resize_maxpool_layer(&l, w, h);
+ }else if(l.type == AVGPOOL){
+ resize_avgpool_layer(&l, w, h);
+ break;
+ }else if(l.type == NORMALIZATION){
+ resize_normalization_layer(&l, w, h);
+ }else if(l.type == COST){
+ resize_cost_layer(&l, inputs);
}else{
error("Cannot resize this type of layer");
}
+ inputs = l.outputs;
+ net->layers[i] = l;
+ w = l.out_w;
+ h = l.out_h;
}
+ //fprintf(stderr, " Done!\n");
return 0;
}
int get_network_output_size(network net)
{
int i;
- for(i = net.n-1; i > 0; --i) if(net.types[i] != COST) break;
- return get_network_output_size_layer(net, i);
+ for(i = net.n-1; i > 0; --i) if(net.layers[i].type != COST) break;
+ return net.layers[i].outputs;
}
int get_network_input_size(network net)
{
- return get_network_input_size_layer(net, 0);
+ return net.layers[0].inputs;
+}
+
+detection_layer get_network_detection_layer(network net)
+{
+ int i;
+ for(i = 0; i < net.n; ++i){
+ if(net.layers[i].type == DETECTION){
+ return net.layers[i];
+ }
+ }
+ fprintf(stderr, "Detection layer not found!!\n");
+ detection_layer l = {0};
+ return l;
}
image get_network_image_layer(network net, int i)
{
- if(net.types[i] == CONVOLUTIONAL){
- convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- return get_convolutional_image(layer);
+ layer l = net.layers[i];
+ if (l.out_w && l.out_h && l.out_c){
+ return float_to_image(l.out_w, l.out_h, l.out_c, l.output);
}
- 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);
- }
- else if(net.types[i] == NORMALIZATION){
- 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);
- }
- return make_empty_image(0,0,0);
+ image def = {0};
+ return def;
}
image get_network_image(network net)
@@ -618,7 +423,8 @@
image m = get_network_image_layer(net, i);
if(m.h != 0) return m;
}
- return make_empty_image(0,0,0);
+ image def = {0};
+ return def;
}
void visualize_network(network net)
@@ -626,16 +432,11 @@
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){
- convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- prev = visualize_convolutional_layer(layer, buff, prev);
- }
- if(net.types[i] == NORMALIZATION){
- normalization_layer layer = *(normalization_layer *)net.layers[i];
- visualize_normalization_layer(layer, buff);
+ layer l = net.layers[i];
+ if(l.type == CONVOLUTIONAL){
+ prev = visualize_convolutional_layer(l, buff, prev);
}
}
}
@@ -650,11 +451,18 @@
float *network_predict(network net, float *input)
{
- #ifdef GPU
+#ifdef GPU
if(gpu_index >= 0) return network_predict_gpu(net, input);
- #endif
+#endif
- forward_network(net, input, 0, 0);
+ network_state state;
+ state.net = net;
+ state.index = 0;
+ state.input = input;
+ state.truth = 0;
+ state.train = 0;
+ state.delta = 0;
+ forward_network(net, state);
float *out = get_network_output(net);
return out;
}
@@ -711,36 +519,9 @@
{
int i,j;
for(i = 0; i < net.n; ++i){
- float *output = 0;
- int n = 0;
- if(net.types[i] == CONVOLUTIONAL){
- convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- output = layer.output;
- image m = get_convolutional_image(layer);
- n = m.h*m.w*m.c;
- }
- else if(net.types[i] == MAXPOOL){
- maxpool_layer layer = *(maxpool_layer *)net.layers[i];
- output = layer.output;
- 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;
- n = layer.outputs;
- }
- else if(net.types[i] == SOFTMAX){
- softmax_layer layer = *(softmax_layer *)net.layers[i];
- output = layer.output;
- n = layer.inputs;
- }
+ layer l = net.layers[i];
+ float *output = l.output;
+ int n = l.outputs;
float mean = mean_array(output, n);
float vari = variance_array(output, n);
fprintf(stderr, "Layer %d - Mean: %f, Variance: %f\n",i,mean, vari);
@@ -784,12 +565,12 @@
return acc;
}
-float *network_accuracies(network net, data d)
+float *network_accuracies(network net, data d, int n)
{
static float acc[2];
matrix guess = network_predict_data(net, d);
- acc[0] = matrix_topk_accuracy(d.y, guess,1);
- acc[1] = matrix_topk_accuracy(d.y, guess,5);
+ acc[0] = matrix_topk_accuracy(d.y, guess, 1);
+ acc[1] = matrix_topk_accuracy(d.y, guess, n);
free_matrix(guess);
return acc;
}
@@ -803,4 +584,17 @@
return acc;
}
-
+void free_network(network net)
+{
+ int i;
+ for(i = 0; i < net.n; ++i){
+ free_layer(net.layers[i]);
+ }
+ free(net.layers);
+ #ifdef GPU
+ if(*net.input_gpu) cuda_free(*net.input_gpu);
+ if(*net.truth_gpu) cuda_free(*net.truth_gpu);
+ if(net.input_gpu) free(net.input_gpu);
+ if(net.truth_gpu) free(net.truth_gpu);
+ #endif
+}
--
Gitblit v1.10.0