mirror/darknet
1
0
Fork 0
mirror of https://github.com/pjreddie/darknet.git synced 2023-08-10 21:13:14 +03:00
Code Issues Projects Releases Wiki Activity

Can validate on imagenet now

Browse Source
This commit is contained in:
Joseph Redmon
2014-11-05 14:49:58 -08:00
parent 2b2441313b
commit b13ad6d5fd
15 changed files with 451 additions and 291 deletions
Download Patch File Download Diff File Expand all files Collapse all files

229
src/network.c
View File

@ -31,150 +31,6 @@ network make_network(int n, int batch)
return net;
}
#ifdef GPU
void forward_network_gpu(network net, cl_mem input, cl_mem truth, int train)
{
//printf("start\n");
int i;
for(i = 0; i < net.n; ++i){
//clock_t time = clock();
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
forward_convolutional_layer_gpu(layer, input);
input = layer.output_cl;
}
else if(net.types[i] == COST){
cost_layer layer = *(cost_layer *)net.layers[i];
forward_cost_layer_gpu(layer, input, truth);
}
else if(net.types[i] == CONNECTED){
connected_layer layer = *(connected_layer *)net.layers[i];
forward_connected_layer_gpu(layer, input);
input = layer.output_cl;
}
else if(net.types[i] == MAXPOOL){
maxpool_layer layer = *(maxpool_layer *)net.layers[i];
forward_maxpool_layer_gpu(layer, input);
input = layer.output_cl;
}
else if(net.types[i] == SOFTMAX){
softmax_layer layer = *(softmax_layer *)net.layers[i];
forward_softmax_layer_gpu(layer, input);
input = layer.output_cl;
}
//printf("%d %f\n", i, sec(clock()-time));
/*
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] == NORMALIZATION){
normalization_layer layer = *(normalization_layer *)net.layers[i];
forward_normalization_layer(layer, input);
input = layer.output;
}
*/
}
}
void backward_network_gpu(network net, cl_mem input)
{
int i;
cl_mem prev_input;
cl_mem prev_delta;
for(i = net.n-1; i >= 0; --i){
//clock_t time = clock();
if(i == 0){
prev_input = input;
prev_delta = 0;
}else{
prev_input = get_network_output_cl_layer(net, i-1);
prev_delta = get_network_delta_cl_layer(net, i-1);
}
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
backward_convolutional_layer_gpu(layer, prev_delta);
}
else if(net.types[i] == COST){
cost_layer layer = *(cost_layer *)net.layers[i];
backward_cost_layer_gpu(layer, prev_input, prev_delta);
}
else if(net.types[i] == CONNECTED){
connected_layer layer = *(connected_layer *)net.layers[i];
backward_connected_layer_gpu(layer, prev_input, prev_delta);
}
else if(net.types[i] == MAXPOOL){
maxpool_layer layer = *(maxpool_layer *)net.layers[i];
backward_maxpool_layer_gpu(layer, prev_delta);
}
else if(net.types[i] == SOFTMAX){
softmax_layer layer = *(softmax_layer *)net.layers[i];
backward_softmax_layer_gpu(layer, prev_delta);
}
//printf("back: %d %f\n", i, sec(clock()-time));
}
}
void update_network_gpu(network net)
{
int i;
for(i = 0; i < net.n; ++i){
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
update_convolutional_layer_gpu(layer);
}
else if(net.types[i] == CONNECTED){
connected_layer layer = *(connected_layer *)net.layers[i];
update_connected_layer_gpu(layer);
}
}
}
cl_mem get_network_output_cl_layer(network net, int i)
{
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
return layer.output_cl;
}
else if(net.types[i] == CONNECTED){
connected_layer layer = *(connected_layer *)net.layers[i];
return layer.output_cl;
}
else if(net.types[i] == MAXPOOL){
maxpool_layer layer = *(maxpool_layer *)net.layers[i];
return layer.output_cl;
}
else if(net.types[i] == SOFTMAX){
softmax_layer layer = *(softmax_layer *)net.layers[i];
return layer.output_cl;
}
return 0;
}
cl_mem get_network_delta_cl_layer(network net, int i)
{
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
return layer.delta_cl;
}
else if(net.types[i] == CONNECTED){
connected_layer layer = *(connected_layer *)net.layers[i];
return layer.delta_cl;
}
else if(net.types[i] == MAXPOOL){
maxpool_layer layer = *(maxpool_layer *)net.layers[i];
return layer.delta_cl;
}
else if(net.types[i] == SOFTMAX){
softmax_layer layer = *(softmax_layer *)net.layers[i];
return layer.delta_cl;
}
return 0;
}
#endif
void forward_network(network net, float *input, float *truth, int train)
{
@ -383,70 +239,6 @@ void backward_network(network net, float *input)
}
#ifdef GPU
float train_network_datum_gpu(network net, float *x, float *y)
{
int x_size = get_network_input_size(net)*net.batch;
int y_size = get_network_output_size(net)*net.batch;
clock_t time = clock();
if(!*net.input_cl){
*net.input_cl = cl_make_array(x, x_size);
*net.truth_cl = cl_make_array(y, y_size);
}else{
cl_write_array(*net.input_cl, x, x_size);
cl_write_array(*net.truth_cl, y, y_size);
}
//printf("trans %f\n", sec(clock()-time));
time = clock();
forward_network_gpu(net, *net.input_cl, *net.truth_cl, 1);
//printf("forw %f\n", sec(clock()-time));
time = clock();
backward_network_gpu(net, *net.input_cl);
//printf("back %f\n", sec(clock()-time));
time = clock();
float error = get_network_cost(net);
update_network_gpu(net);
//printf("updt %f\n", sec(clock()-time));
time = clock();
return error;
}
float train_network_sgd_gpu(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_random_batch(d, batch, X, y);
float err = train_network_datum_gpu(net, X, y);
sum += err;
}
free(X);
free(y);
return (float)sum/(n*batch);
}
float train_network_data_gpu(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_gpu(net, X, y);
sum += err;
}
free(X);
free(y);
return (float)sum/(n*batch);
}
#endif
float train_network_datum(network net, float *x, float *y)
@ -477,6 +269,7 @@ float train_network_sgd(network net, data d, int n)
free(y);
return (float)sum/(n*batch);
}
float train_network_batch(network net, data d, int n)
{
int i,j;
@ -496,6 +289,23 @@ float train_network_batch(network net, data d, int n)
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)
{
@ -687,6 +497,7 @@ void top_predictions(network net, int n, int *index)
}
}
float *network_predict(network net, float *input)
{
forward_network(net, input, 0, 0);
@ -724,7 +535,7 @@ 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;