From 481b57a96a9ef29b112caec1bb3e17ffb043ceae Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Sun, 25 Sep 2016 06:12:54 +0000
Subject: [PATCH] So I have this new programming paradigm.......
---
src/network.c | 142 ++++++++++++-----------------------------------
1 files changed, 37 insertions(+), 105 deletions(-)
diff --git a/src/network.c b/src/network.c
index 88b7085..01b7962 100644
--- a/src/network.c
+++ b/src/network.c
@@ -1,5 +1,6 @@
#include <stdio.h>
#include <time.h>
+#include <assert.h>
#include "network.h"
#include "image.h"
#include "data.h"
@@ -14,11 +15,12 @@
#include "local_layer.h"
#include "convolutional_layer.h"
#include "activation_layer.h"
-#include "deconvolutional_layer.h"
#include "detection_layer.h"
+#include "region_layer.h"
#include "normalization_layer.h"
#include "batchnorm_layer.h"
#include "maxpool_layer.h"
+#include "reorg_layer.h"
#include "avgpool_layer.h"
#include "cost_layer.h"
#include "softmax_layer.h"
@@ -64,6 +66,7 @@
case EXP:
return net.learning_rate * pow(net.gamma, batch_num);
case POLY:
+ if (batch_num < net.burn_in) return net.learning_rate * pow((float)batch_num / net.burn_in, net.power);
return net.learning_rate * pow(1 - (float)batch_num / net.max_batches, net.power);
case RANDOM:
return net.learning_rate * pow(rand_uniform(0,1), net.power);
@@ -96,12 +99,16 @@
return "crnn";
case MAXPOOL:
return "maxpool";
+ case REORG:
+ return "reorg";
case AVGPOOL:
return "avgpool";
case SOFTMAX:
return "softmax";
case DETECTION:
return "detection";
+ case REGION:
+ return "region";
case DROPOUT:
return "dropout";
case CROP:
@@ -145,45 +152,7 @@
if(l.delta){
scal_cpu(l.outputs * l.batch, 0, l.delta, 1);
}
- if(l.type == CONVOLUTIONAL){
- forward_convolutional_layer(l, state);
- } else if(l.type == DECONVOLUTIONAL){
- forward_deconvolutional_layer(l, state);
- } else if(l.type == ACTIVE){
- forward_activation_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 == BATCHNORM){
- forward_batchnorm_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 == RNN){
- forward_rnn_layer(l, state);
- } else if(l.type == GRU){
- forward_gru_layer(l, state);
- } else if(l.type == CRNN){
- forward_crnn_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);
- }
+ l.forward(l, state);
state.input = l.output;
}
}
@@ -195,29 +164,17 @@
float rate = get_current_rate(net);
for(i = 0; i < net.n; ++i){
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 == RNN){
- update_rnn_layer(l, update_batch, rate, net.momentum, net.decay);
- } else if(l.type == GRU){
- update_gru_layer(l, update_batch, rate, net.momentum, net.decay);
- } else if(l.type == CRNN){
- update_crnn_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);
+ if(l.update){
+ l.update(l, update_batch, rate, net.momentum, net.decay);
}
}
}
float *get_network_output(network net)
{
- #ifdef GPU
- return get_network_output_gpu(net);
- #endif
+#ifdef GPU
+ if (gpu_index >= 0) return get_network_output_gpu(net);
+#endif
int i;
for(i = net.n-1; i > 0; --i) if(net.layers[i].type != COST) break;
return net.layers[i].output;
@@ -229,11 +186,7 @@
float sum = 0;
int count = 0;
for(i = 0; i < net.n; ++i){
- if(net.layers[i].type == COST){
- sum += net.layers[i].cost[0];
- ++count;
- }
- if(net.layers[i].type == DETECTION){
+ if(net.layers[i].cost){
sum += net.layers[i].cost[0];
++count;
}
@@ -253,6 +206,7 @@
int i;
float *original_input = state.input;
float *original_delta = state.delta;
+ state.workspace = net.workspace;
for(i = net.n-1; i >= 0; --i){
state.index = i;
if(i == 0){
@@ -264,53 +218,17 @@
state.delta = 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 == ACTIVE){
- backward_activation_layer(l, state);
- } else if(l.type == NORMALIZATION){
- backward_normalization_layer(l, state);
- } else if(l.type == BATCHNORM){
- backward_batchnorm_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 == RNN){
- backward_rnn_layer(l, state);
- } else if(l.type == GRU){
- backward_gru_layer(l, state);
- } else if(l.type == CRNN){
- backward_crnn_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);
- }
+ l.backward(l, state);
}
}
float train_network_datum(network net, float *x, float *y)
{
- *net.seen += net.batch;
#ifdef GPU
if(gpu_index >= 0) return train_network_datum_gpu(net, x, y);
#endif
network_state state;
+ *net.seen += net.batch;
state.index = 0;
state.net = net;
state.input = x;
@@ -344,6 +262,7 @@
float train_network(network net, data d)
{
+ assert(d.X.rows % net.batch == 0);
int batch = net.batch;
int n = d.X.rows / batch;
float *X = calloc(batch*d.X.cols, sizeof(float));
@@ -361,6 +280,7 @@
return (float)sum/(n*batch);
}
+
float train_network_batch(network net, data d, int n)
{
int i,j;
@@ -391,6 +311,11 @@
int i;
for(i = 0; i < net->n; ++i){
net->layers[i].batch = b;
+#ifdef CUDNN
+ if(net->layers[i].type == CONVOLUTIONAL){
+ cudnn_convolutional_setup(net->layers + i);
+ }
+#endif
}
}
@@ -402,7 +327,7 @@
net->h = h;
int inputs = 0;
size_t workspace_size = 0;
- //fprintf(stderr, "Resizing to %d x %d...", w, h);
+ //fprintf(stderr, "Resizing to %d x %d...\n", w, h);
//fflush(stderr);
for (i = 0; i < net->n; ++i){
layer l = net->layers[i];
@@ -412,6 +337,8 @@
resize_crop_layer(&l, w, h);
}else if(l.type == MAXPOOL){
resize_maxpool_layer(&l, w, h);
+ }else if(l.type == REORG){
+ resize_reorg_layer(&l, w, h);
}else if(l.type == AVGPOOL){
resize_avgpool_layer(&l, w, h);
}else if(l.type == NORMALIZATION){
@@ -429,11 +356,16 @@
if(l.type == AVGPOOL) break;
}
#ifdef GPU
+ if(gpu_index >= 0){
cuda_free(net->workspace);
net->workspace = cuda_make_array(0, (workspace_size-1)/sizeof(float)+1);
-#else
+ }else {
free(net->workspace);
- net->workspace = calloc(1, (workspace_size-1)/sizeof(float)+1);
+ net->workspace = calloc(1, workspace_size);
+ }
+#else
+ free(net->workspace);
+ net->workspace = calloc(1, workspace_size);
#endif
//fprintf(stderr, " Done!\n");
return 0;
@@ -649,10 +581,10 @@
free_layer(net.layers[i]);
}
free(net.layers);
- #ifdef GPU
+#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
+#endif
}
--
Gitblit v1.10.0