From cf32e7e9b843560eb7ec3ed16e5b19f0f7156724 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@burninator.cs.washington.edu>
Date: Sat, 25 Jun 2016 23:12:00 +0000
Subject: [PATCH] colors
---
src/network.c | 84 +++++++++++++++++++++++++++++++++++++++++-
1 files changed, 82 insertions(+), 2 deletions(-)
diff --git a/src/network.c b/src/network.c
index d9585c4..a9e5027 100644
--- a/src/network.c
+++ b/src/network.c
@@ -8,17 +8,23 @@
#include "crop_layer.h"
#include "connected_layer.h"
+#include "gru_layer.h"
+#include "rnn_layer.h"
+#include "crnn_layer.h"
#include "local_layer.h"
#include "convolutional_layer.h"
+#include "activation_layer.h"
#include "deconvolutional_layer.h"
#include "detection_layer.h"
#include "normalization_layer.h"
+#include "batchnorm_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)
{
@@ -58,7 +64,10 @@
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);
case SIG:
return net.learning_rate * (1./(1.+exp(net.gamma*(batch_num - net.step))));
default:
@@ -72,12 +81,20 @@
switch(a){
case CONVOLUTIONAL:
return "convolutional";
+ case ACTIVE:
+ return "activation";
case LOCAL:
return "local";
case DECONVOLUTIONAL:
return "deconvolutional";
case CONNECTED:
return "connected";
+ case RNN:
+ return "rnn";
+ case GRU:
+ return "gru";
+ case CRNN:
+ return "crnn";
case MAXPOOL:
return "maxpool";
case AVGPOOL:
@@ -94,8 +111,12 @@
return "cost";
case ROUTE:
return "route";
+ case SHORTCUT:
+ return "shortcut";
case NORMALIZATION:
return "normalization";
+ case BATCHNORM:
+ return "batchnorm";
default:
break;
}
@@ -117,8 +138,10 @@
void forward_network(network net, network_state state)
{
+ state.workspace = net.workspace;
int i;
for(i = 0; i < net.n; ++i){
+ state.index = i;
layer l = net.layers[i];
if(l.delta){
scal_cpu(l.outputs * l.batch, 0, l.delta, 1);
@@ -127,14 +150,24 @@
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){
@@ -149,6 +182,8 @@
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);
}
state.input = l.output;
}
@@ -167,6 +202,12 @@
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);
}
@@ -175,6 +216,9 @@
float *get_network_output(network net)
{
+ #ifdef GPU
+ 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;
@@ -187,7 +231,7 @@
int count = 0;
for(i = 0; i < net.n; ++i){
if(net.layers[i].type == COST){
- sum += net.layers[i].output[0];
+ sum += net.layers[i].cost[0];
++count;
}
if(net.layers[i].type == DETECTION){
@@ -210,7 +254,9 @@
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){
state.input = original_input;
state.delta = original_delta;
@@ -224,8 +270,12 @@
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){
@@ -238,12 +288,20 @@
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);
}
}
}
@@ -255,6 +313,8 @@
if(gpu_index >= 0) return train_network_datum_gpu(net, x, y);
#endif
network_state state;
+ state.index = 0;
+ state.net = net;
state.input = x;
state.delta = 0;
state.truth = y;
@@ -307,6 +367,8 @@
{
int i,j;
network_state state;
+ state.index = 0;
+ state.net = net;
state.train = 1;
state.delta = 0;
float sum = 0;
@@ -331,6 +393,11 @@
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
}
}
@@ -341,17 +408,19 @@
net->w = w;
net->h = h;
int inputs = 0;
+ size_t workspace_size = 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 == CROP){
+ resize_crop_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){
@@ -359,11 +428,20 @@
}else{
error("Cannot resize this type of layer");
}
+ if(l.workspace_size > workspace_size) workspace_size = l.workspace_size;
inputs = l.outputs;
net->layers[i] = l;
w = l.out_w;
h = l.out_h;
+ if(l.type == AVGPOOL) break;
}
+#ifdef GPU
+ cuda_free(net->workspace);
+ net->workspace = cuda_make_array(0, (workspace_size-1)/sizeof(float)+1);
+#else
+ free(net->workspace);
+ net->workspace = calloc(1, workspace_size);
+#endif
//fprintf(stderr, " Done!\n");
return 0;
}
@@ -443,6 +521,8 @@
#endif
network_state state;
+ state.net = net;
+ state.index = 0;
state.input = input;
state.truth = 0;
state.train = 0;
--
Gitblit v1.10.0