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

it's raining really hard outside :-( :rain: :storm: ☁️

Browse Source
This commit is contained in:
Joseph Redmon 2017-10-17 11:41:34 -07:00
parent 532c6e1481
commit cd5d393b46
27 changed files with 1340 additions and 1669 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Makefile
View File

@ -58,10 +58,10 @@ LDFLAGS+= -lcudnn
endif
OBJ=gemm.o utils.o cuda.o deconvolutional_layer.o convolutional_layer.o list.o image.o activations.o im2col.o col2im.o blas.o crop_layer.o dropout_layer.o maxpool_layer.o softmax_layer.o data.o matrix.o network.o connected_layer.o cost_layer.o parser.o option_list.o detection_layer.o route_layer.o box.o normalization_layer.o avgpool_layer.o layer.o local_layer.o shortcut_layer.o activation_layer.o rnn_layer.o gru_layer.o crnn_layer.o demo.o batchnorm_layer.o region_layer.o reorg_layer.o tree.o lstm_layer.o
EXECOBJA=captcha.o lsd.o super.o voxel.o art.o tag.o cifar.o go.o rnn.o rnn_vid.o compare.o segmenter.o regressor.o classifier.o coco.o dice.o yolo.o detector.o writing.o nightmare.o swag.o darknet.o
EXECOBJA=captcha.o lsd.o super.o art.o tag.o cifar.o go.o rnn.o segmenter.o regressor.o classifier.o coco.o yolo.o detector.o nightmare.o attention.o darknet.o
ifeq ($(GPU), 1)
LDFLAGS+= -lstdc++
OBJ+=convolutional_kernels.o deconvolutional_kernels.o activation_kernels.o im2col_kernels.o col2im_kernels.o blas_kernels.o crop_layer_kernels.o dropout_layer_kernels.o maxpool_layer_kernels.o network_kernels.o avgpool_layer_kernels.o
OBJ+=convolutional_kernels.o deconvolutional_kernels.o activation_kernels.o im2col_kernels.o col2im_kernels.o blas_kernels.o crop_layer_kernels.o dropout_layer_kernels.o maxpool_layer_kernels.o avgpool_layer_kernels.o
endif
EXECOBJ = $(addprefix $(OBJDIR), $(EXECOBJA))

2
cfg/darknet.cfg
View File

@ -1,6 +1,6 @@
[net]
# Train
batch=128
batch=1
subdivisions=1
# Test
# batch=1

9
examples/art.c
View File

@ -5,11 +5,8 @@
void demo_art(char *cfgfile, char *weightfile, int cam_index)
{
#ifdef OPENCV
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
CvCapture * cap;
@ -26,7 +23,7 @@ void demo_art(char *cfgfile, char *weightfile, int cam_index)
while(1){
image in = get_image_from_stream(cap);
image in_s = resize_image(in, net.w, net.h);
image in_s = resize_image(in, net->w, net->h);
show_image(in, window);
float *p = network_predict(net, in_s.data);

51
examples/captcha.c
View File

@ -30,13 +30,10 @@ void train_captcha(char *cfgfile, char *weightfile)
float avg_loss = -1;
char *base = basecfg(cfgfile);
printf("%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = 1024;
int i = *net.seen/imgs;
int i = *net->seen/imgs;
int solved = 1;
list *plist;
char **labels = get_labels("/data/captcha/reimgs.labels.list");
@ -53,8 +50,8 @@ void train_captcha(char *cfgfile, char *weightfile)
data buffer;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.classes = 26;
args.n = imgs;
@ -83,7 +80,7 @@ void train_captcha(char *cfgfile, char *weightfile)
float loss = train_network(net, train);
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%d: %f, %f avg, %lf seconds, %ld images\n", i, loss, avg_loss, sec(clock()-time), *net.seen);
printf("%d: %f, %f avg, %lf seconds, %ld images\n", i, loss, avg_loss, sec(clock()-time), *net->seen);
free_data(train);
if(i%100==0){
char buff[256];
@ -95,11 +92,8 @@ void train_captcha(char *cfgfile, char *weightfile)
void test_captcha(char *cfgfile, char *weightfile, char *filename)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
int i = 0;
char **names = get_labels("/data/captcha/reimgs.labels.list");
@ -116,7 +110,7 @@ void test_captcha(char *cfgfile, char *weightfile, char *filename)
if(!input) return;
strtok(input, "\n");
}
image im = load_image_color(input, net.w, net.h);
image im = load_image_color(input, net->w, net->h);
float *X = im.data;
float *predictions = network_predict(net, X);
top_predictions(net, 26, indexes);
@ -136,21 +130,18 @@ void test_captcha(char *cfgfile, char *weightfile, char *filename)
void valid_captcha(char *cfgfile, char *weightfile, char *filename)
{
char **labels = get_labels("/data/captcha/reimgs.labels.list");
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
list *plist = get_paths("/data/captcha/reimgs.fg.list");
char **paths = (char **)list_to_array(plist);
int N = plist->size;
int outputs = net.outputs;
int outputs = net->outputs;
set_batch_network(&net, 1);
set_batch_network(net, 1);
srand(2222222);
int i, j;
for(i = 0; i < N; ++i){
if (i%100 == 0) fprintf(stderr, "%d\n", i);
image im = load_image_color(paths[i], net.w, net.h);
image im = load_image_color(paths[i], net->w, net->h);
float *X = im.data;
float *predictions = network_predict(net, X);
//printf("%s: Predicted in %f seconds.\n", input, sec(clock()-time));
@ -185,9 +176,9 @@ void valid_captcha(char *cfgfile, char *weightfile, char *filename)
if(weightfile){
load_weights(&net, weightfile);
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = 1024;
int i = net.seen/imgs;
int i = net->seen/imgs;
list *plist = get_paths("/data/captcha/train.auto5");
char **paths = (char **)list_to_array(plist);
printf("%d\n", plist->size);
@ -201,10 +192,10 @@ void valid_captcha(char *cfgfile, char *weightfile, char *filename)
printf("Loaded: %lf seconds\n", sec(clock()-time));
time=clock();
float loss = train_network(net, train);
net.seen += imgs;
net->seen += imgs;
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%d: %f, %f avg, %lf seconds, %d images\n", i, loss, avg_loss, sec(clock()-time), net.seen);
printf("%d: %f, %f avg, %lf seconds, %d images\n", i, loss, avg_loss, sec(clock()-time), net->seen);
free_data(train);
if(i%10==0){
char buff[256];
@ -251,9 +242,9 @@ network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = 1024;
int i = net.seen/imgs;
int i = net->seen/imgs;
list *plist = get_paths("/data/captcha/encode.list");
char **paths = (char **)list_to_array(plist);
printf("%d\n", plist->size);
@ -266,10 +257,10 @@ while(1){
printf("Loaded: %lf seconds\n", sec(clock()-time));
time=clock();
float loss = train_network(net, train);
net.seen += imgs;
net->seen += imgs;
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%d: %f, %f avg, %lf seconds, %d images\n", i, loss, avg_loss, sec(clock()-time), net.seen);
printf("%d: %f, %f avg, %lf seconds, %d images\n", i, loss, avg_loss, sec(clock()-time), net->seen);
free_matrix(train.X);
if(i%100==0){
char buff[256];

58
examples/cifar.c
View File

@ -6,28 +6,25 @@ void train_cifar(char *cfgfile, char *weightfile)
float avg_loss = -1;
char *base = basecfg(cfgfile);
printf("%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
char *backup_directory = "/home/pjreddie/backup/";
int classes = 10;
int N = 50000;
char **labels = get_labels("data/cifar/labels.txt");
int epoch = (*net.seen)/N;
int epoch = (*net->seen)/N;
data train = load_all_cifar10();
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
while(get_current_batch(net) < net->max_batches || net->max_batches == 0){
clock_t time=clock();
float loss = train_network_sgd(net, train, 1);
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.95 + loss*.05;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net->seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net->seen);
if(*net->seen/N > epoch){
epoch = *net->seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
save_weights(net, buff);
@ -54,18 +51,15 @@ void train_cifar_distill(char *cfgfile, char *weightfile)
float avg_loss = -1;
char *base = basecfg(cfgfile);
printf("%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
char *backup_directory = "/home/pjreddie/backup/";
int classes = 10;
int N = 50000;
char **labels = get_labels("data/cifar/labels.txt");
int epoch = (*net.seen)/N;
int epoch = (*net->seen)/N;
data train = load_all_cifar10();
matrix soft = csv_to_matrix("results/ensemble.csv");
@ -75,15 +69,15 @@ void train_cifar_distill(char *cfgfile, char *weightfile)
scale_matrix(train.y, 1. - weight);
matrix_add_matrix(soft, train.y);
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
while(get_current_batch(net) < net->max_batches || net->max_batches == 0){
clock_t time=clock();
float loss = train_network_sgd(net, train, 1);
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.95 + loss*.05;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net->seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net->seen);
if(*net->seen/N > epoch){
epoch = *net->seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
save_weights(net, buff);
@ -106,11 +100,8 @@ void train_cifar_distill(char *cfgfile, char *weightfile)
void test_cifar_multi(char *filename, char *weightfile)
{
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(filename, weightfile, 0);
set_batch_network(net, 1);
srand(time(0));
float avg_acc = 0;
@ -138,10 +129,7 @@ void test_cifar_multi(char *filename, char *weightfile)
void test_cifar(char *filename, char *weightfile)
{
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
srand(time(0));
clock_t time;
@ -182,10 +170,7 @@ char *labels[] = {"airplane","automobile","bird","cat","deer","dog","frog","hors
void test_cifar_csv(char *filename, char *weightfile)
{
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
srand(time(0));
data test = load_cifar10_data("data/cifar/cifar-10-batches-bin/test_batch.bin");
@ -207,12 +192,9 @@ void test_cifar_csv(char *filename, char *weightfile)
free_data(test);
}
void test_cifar_csvtrain(char *filename, char *weightfile)
void test_cifar_csvtrain(char *cfg, char *weights)
{
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfg, weights, 0);
srand(time(0));
data test = load_all_cifar10();

329
examples/classifier.c
View File

@ -23,7 +23,7 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
char *base = basecfg(cfgfile);
printf("%s\n", base);
printf("%d\n", ngpus);
network *nets = calloc(ngpus, sizeof(network));
network **nets = calloc(ngpus, sizeof(network*));
srand(time(0));
int seed = rand();
@ -33,14 +33,14 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
cuda_set_device(gpus[i]);
#endif
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i].learning_rate *= ngpus;
nets[i]->learning_rate *= ngpus;
}
srand(time(0));
network net = nets[0];
network *net = nets[0];
int imgs = net.batch * net.subdivisions * ngpus;
int imgs = net->batch * net->subdivisions * ngpus;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
list *options = read_data_cfg(datacfg);
char *backup_directory = option_find_str(options, "backup", "/backup/");
@ -56,19 +56,20 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
double time;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.threads = 32;
args.hierarchy = net.hierarchy;
args.hierarchy = net->hierarchy;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.size = net.w;
args.min = net->min_ratio*net->w;
args.max = net->max_ratio*net->w;
printf("%d %d\n", args.min, args.max);
args.angle = net->angle;
args.aspect = net->aspect;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
args.size = net->w;
args.paths = paths;
args.classes = classes;
@ -83,8 +84,32 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
args.d = &buffer;
load_thread = load_data(args);
int epoch = (*net.seen)/N;
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
int count = 0;
int epoch = (*net->seen)/N;
while(get_current_batch(net) < net->max_batches || net->max_batches == 0){
if(net->random && count++%40 == 0){
printf("Resizing\n");
int dim = (rand() % 11 + 4) * 32;
//if (get_current_batch(net)+200 > net->max_batches) dim = 608;
//int dim = (rand() % 4 + 16) * 32;
printf("%d\n", dim);
args.w = dim;
args.h = dim;
args.size = dim;
args.min = net->min_ratio*dim;
args.max = net->max_ratio*dim;
printf("%d %d\n", args.min, args.max);
pthread_join(load_thread, 0);
train = buffer;
free_data(train);
load_thread = load_data(args);
for(i = 0; i < ngpus; ++i){
resize_network(nets[i], dim, dim);
}
net = nets[0];
}
time = what_time_is_it_now();
pthread_join(load_thread, 0);
@ -106,10 +131,10 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
#endif
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), what_time_is_it_now()-time, *net.seen);
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net->seen)/N, loss, avg_loss, get_current_rate(net), what_time_is_it_now()-time, *net->seen);
free_data(train);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
if(*net->seen/N > epoch){
epoch = *net->seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
save_weights(net, buff);
@ -132,124 +157,10 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
free(base);
}
/*
void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int clear)
{
srand(time(0));
float avg_loss = -1;
char *base = basecfg(cfgfile);
printf("%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
if(clear) *net.seen = 0;
int imgs = net.batch * net.subdivisions;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
list *options = read_data_cfg(datacfg);
char *backup_directory = option_find_str(options, "backup", "/backup/");
char *label_list = option_find_str(options, "labels", "data/labels.list");
char *train_list = option_find_str(options, "train", "data/train.list");
int classes = option_find_int(options, "classes", 2);
char **labels = get_labels(label_list);
list *plist = get_paths(train_list);
char **paths = (char **)list_to_array(plist);
printf("%d\n", plist->size);
int N = plist->size;
clock_t time;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.threads = 8;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.size = net.w;
args.hierarchy = net.hierarchy;
args.paths = paths;
args.classes = classes;
args.n = imgs;
args.m = N;
args.labels = labels;
args.type = CLASSIFICATION_DATA;
data train;
data buffer;
pthread_t load_thread;
args.d = &buffer;
load_thread = load_data(args);
int epoch = (*net.seen)/N;
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
time=clock();
pthread_join(load_thread, 0);
train = buffer;
load_thread = load_data(args);
printf("Loaded: %lf seconds\n", sec(clock()-time));
time=clock();
#ifdef OPENCV
if(0){
int u;
for(u = 0; u < imgs; ++u){
image im = float_to_image(net.w, net.h, 3, train.X.vals[u]);
show_image(im, "loaded");
cvWaitKey(0);
}
}
#endif
float loss = train_network(net, train);
free_data(train);
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%d, %.3f: %f, %f avg, %f rate, %lf seconds, %d images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
save_weights(net, buff);
}
if(get_current_batch(net)%100 == 0){
char buff[256];
sprintf(buff, "%s/%s.backup",backup_directory,base);
save_weights(net, buff);
}
}
char buff[256];
sprintf(buff, "%s/%s.weights", backup_directory, base);
save_weights(net, buff);
free_network(net);
free_ptrs((void**)labels, classes);
free_ptrs((void**)paths, plist->size);
free_list(plist);
free(base);
}
*/
void validate_classifier_crop(char *datacfg, char *filename, char *weightfile)
{
int i = 0;
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
srand(time(0));
list *options = read_data_cfg(datacfg);
@ -275,8 +186,8 @@ void validate_classifier_crop(char *datacfg, char *filename, char *weightfile)
data val, buffer;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.classes = classes;
@ -313,11 +224,8 @@ void validate_classifier_crop(char *datacfg, char *filename, char *weightfile)
void validate_classifier_10(char *datacfg, char *filename, char *weightfile)
{
int i, j;
network net = parse_network_cfg(filename);
set_batch_network(&net, 1);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
set_batch_network(net, 1);
srand(time(0));
list *options = read_data_cfg(datacfg);
@ -347,8 +255,8 @@ void validate_classifier_10(char *datacfg, char *filename, char *weightfile)
break;
}
}
int w = net.w;
int h = net.h;
int w = net->w;
int h = net->h;
int shift = 32;
image im = load_image_color(paths[i], w+shift, h+shift);
image images[10];
@ -366,7 +274,7 @@ void validate_classifier_10(char *datacfg, char *filename, char *weightfile)
float *pred = calloc(classes, sizeof(float));
for(j = 0; j < 10; ++j){
float *p = network_predict(net, images[j].data);
if(net.hierarchy) hierarchy_predictions(p, net.outputs, net.hierarchy, 1, 1);
if(net->hierarchy) hierarchy_predictions(p, net->outputs, net->hierarchy, 1, 1);
axpy_cpu(classes, 1, p, 1, pred, 1);
free_image(images[j]);
}
@ -385,11 +293,8 @@ void validate_classifier_10(char *datacfg, char *filename, char *weightfile)
void validate_classifier_full(char *datacfg, char *filename, char *weightfile)
{
int i, j;
network net = parse_network_cfg(filename);
set_batch_network(&net, 1);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
set_batch_network(net, 1);
srand(time(0));
list *options = read_data_cfg(datacfg);
@ -410,7 +315,7 @@ void validate_classifier_full(char *datacfg, char *filename, char *weightfile)
float avg_topk = 0;
int *indexes = calloc(topk, sizeof(int));
int size = net.w;
int size = net->w;
for(i = 0; i < m; ++i){
int class = -1;
char *path = paths[i];
@ -422,12 +327,12 @@ void validate_classifier_full(char *datacfg, char *filename, char *weightfile)
}
image im = load_image_color(paths[i], 0, 0);
image resized = resize_min(im, size);
resize_network(&net, resized.w, resized.h);
resize_network(net, resized.w, resized.h);
//show_image(im, "orig");
//show_image(crop, "cropped");
//cvWaitKey(0);
float *pred = network_predict(net, resized.data);
if(net.hierarchy) hierarchy_predictions(pred, net.outputs, net.hierarchy, 1, 1);
if(net->hierarchy) hierarchy_predictions(pred, net->outputs, net->hierarchy, 1, 1);
free_image(im);
free_image(resized);
@ -446,18 +351,15 @@ void validate_classifier_full(char *datacfg, char *filename, char *weightfile)
void validate_classifier_single(char *datacfg, char *filename, char *weightfile)
{
int i, j;
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(filename, weightfile, 0);
set_batch_network(net, 1);
srand(time(0));
list *options = read_data_cfg(datacfg);
char *label_list = option_find_str(options, "labels", "data/labels.list");
char *leaf_list = option_find_str(options, "leaves", 0);
if(leaf_list) change_leaves(net.hierarchy, leaf_list);
if(leaf_list) change_leaves(net->hierarchy, leaf_list);
char *valid_list = option_find_str(options, "valid", "data/train.list");
int classes = option_find_int(options, "classes", 2);
int topk = option_find_int(options, "top", 1);
@ -483,13 +385,13 @@ void validate_classifier_single(char *datacfg, char *filename, char *weightfile)
}
}
image im = load_image_color(paths[i], 0, 0);
image resized = resize_min(im, net.w);
image crop = crop_image(resized, (resized.w - net.w)/2, (resized.h - net.h)/2, net.w, net.h);
image resized = resize_min(im, net->w);
image crop = crop_image(resized, (resized.w - net->w)/2, (resized.h - net->h)/2, net->w, net->h);
//show_image(im, "orig");
//show_image(crop, "cropped");
//cvWaitKey(0);
float *pred = network_predict(net, crop.data);
if(net.hierarchy) hierarchy_predictions(pred, net.outputs, net.hierarchy, 1, 1);
if(net->hierarchy) hierarchy_predictions(pred, net->outputs, net->hierarchy, 1, 1);
if(resized.data != im.data) free_image(resized);
free_image(im);
@ -505,14 +407,11 @@ void validate_classifier_single(char *datacfg, char *filename, char *weightfile)
}
}
void validate_classifier_multi(char *datacfg, char *filename, char *weightfile)
void validate_classifier_multi(char *datacfg, char *cfg, char *weights)
{
int i, j;
network net = parse_network_cfg(filename);
set_batch_network(&net, 1);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
srand(time(0));
list *options = read_data_cfg(datacfg);
@ -524,7 +423,8 @@ void validate_classifier_multi(char *datacfg, char *filename, char *weightfile)
char **labels = get_labels(label_list);
list *plist = get_paths(valid_list);
int scales[] = {224, 288, 320, 352, 384};
//int scales[] = {224, 288, 320, 352, 384};
int scales[] = {224, 256, 288, 320};
int nscales = sizeof(scales)/sizeof(scales[0]);
char **paths = (char **)list_to_array(plist);
@ -548,9 +448,9 @@ void validate_classifier_multi(char *datacfg, char *filename, char *weightfile)
image im = load_image_color(paths[i], 0, 0);
for(j = 0; j < nscales; ++j){
image r = resize_min(im, scales[j]);
resize_network(&net, r.w, r.h);
resize_network(net, r.w, r.h);
float *p = network_predict(net, r.data);
if(net.hierarchy) hierarchy_predictions(p, net.outputs, net.hierarchy, 1 , 1);
if(net->hierarchy) hierarchy_predictions(p, net->outputs, net->hierarchy, 1 , 1);
axpy_cpu(classes, 1, p, 1, pred, 1);
flip_image(r);
p = network_predict(net, r.data);
@ -571,11 +471,8 @@ void validate_classifier_multi(char *datacfg, char *filename, char *weightfile)
void try_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filename, int layer_num)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
list *options = read_data_cfg(datacfg);
@ -616,7 +513,7 @@ void try_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filena
time=clock();
float *predictions = network_predict(net, X);
layer l = net.layers[layer_num];
layer l = net->layers[layer_num];
for(i = 0; i < l.c; ++i){
if(l.rolling_mean) printf("%f %f %f\n", l.rolling_mean[i], l.rolling_variance[i], l.scales[i]);
}
@ -652,11 +549,8 @@ void try_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filena
void predict_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filename, int top)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
list *options = read_data_cfg(datacfg);
@ -682,19 +576,19 @@ void predict_classifier(char *datacfg, char *cfgfile, char *weightfile, char *fi
strtok(input, "\n");
}
image im = load_image_color(input, 0, 0);
image r = letterbox_image(im, net.w, net.h);
//resize_network(&net, r.w, r.h);
image r = letterbox_image(im, net->w, net->h);
//resize_network(net, r.w, r.h);
//printf("%d %d\n", r.w, r.h);
float *X = r.data;
time=clock();
float *predictions = network_predict(net, X);
if(net.hierarchy) hierarchy_predictions(predictions, net.outputs, net.hierarchy, 1, 1);
top_k(predictions, net.outputs, top, indexes);
if(net->hierarchy) hierarchy_predictions(predictions, net->outputs, net->hierarchy, 1, 1);
top_k(predictions, net->outputs, top, indexes);
fprintf(stderr, "%s: Predicted in %f seconds.\n", input, sec(clock()-time));
for(i = 0; i < top; ++i){
int index = indexes[i];
//if(net.hierarchy) printf("%d, %s: %f, parent: %s \n",index, names[index], predictions[index], (net.hierarchy->parent[index] >= 0) ? names[net.hierarchy->parent[index]] : "Root");
//if(net->hierarchy) printf("%d, %s: %f, parent: %s \n",index, names[index], predictions[index], (net->hierarchy->parent[index] >= 0) ? names[net->hierarchy->parent[index]] : "Root");
//else printf("%s: %f\n",names[index], predictions[index]);
printf("%5.2f%%: %s\n", predictions[index]*100, names[index]);
}
@ -708,11 +602,8 @@ void predict_classifier(char *datacfg, char *cfgfile, char *weightfile, char *fi
void label_classifier(char *datacfg, char *filename, char *weightfile)
{
int i;
network net = parse_network_cfg(filename);
set_batch_network(&net, 1);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
set_batch_network(net, 1);
srand(time(0));
list *options = read_data_cfg(datacfg);
@ -730,8 +621,8 @@ void label_classifier(char *datacfg, char *filename, char *weightfile)
for(i = 0; i < m; ++i){
image im = load_image_color(paths[i], 0, 0);
image resized = resize_min(im, net.w);
image crop = crop_image(resized, (resized.w - net.w)/2, (resized.h - net.h)/2, net.w, net.h);
image resized = resize_min(im, net->w);
image crop = crop_image(resized, (resized.w - net->w)/2, (resized.h - net->h)/2, net->w, net->h);
float *pred = network_predict(net, crop.data);
if(resized.data != im.data) free_image(resized);
@ -747,10 +638,7 @@ void label_classifier(char *datacfg, char *filename, char *weightfile)
void test_classifier(char *datacfg, char *cfgfile, char *weightfile, int target_layer)
{
int curr = 0;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
srand(time(0));
list *options = read_data_cfg(datacfg);
@ -769,18 +657,18 @@ void test_classifier(char *datacfg, char *cfgfile, char *weightfile, int target_
data val, buffer;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.classes = classes;
args.n = net.batch;
args.n = net->batch;
args.m = 0;
args.labels = 0;
args.d = &buffer;
args.type = OLD_CLASSIFICATION_DATA;
pthread_t load_thread = load_data_in_thread(args);
for(curr = net.batch; curr < m; curr += net.batch){
for(curr = net->batch; curr < m; curr += net->batch){
time=clock();
pthread_join(load_thread, 0);
@ -788,7 +676,7 @@ void test_classifier(char *datacfg, char *cfgfile, char *weightfile, int target_
if(curr < m){
args.paths = paths + curr;
if (curr + net.batch > m) args.n = m - curr;
if (curr + net->batch > m) args.n = m - curr;
load_thread = load_data_in_thread(args);
}
fprintf(stderr, "Loaded: %d images in %lf seconds\n", val.X.rows, sec(clock()-time));
@ -798,11 +686,11 @@ void test_classifier(char *datacfg, char *cfgfile, char *weightfile, int target_
int i, j;
if (target_layer >= 0){
//layer l = net.layers[target_layer];
//layer l = net->layers[target_layer];
}
for(i = 0; i < pred.rows; ++i){
printf("%s", paths[curr-net.batch+i]);
printf("%s", paths[curr-net->batch+i]);
for(j = 0; j < pred.cols; ++j){
printf("\t%g", pred.vals[i][j]);
}
@ -824,11 +712,8 @@ void threat_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_i
float roll = .2;
printf("Classifier Demo\n");
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
list *options = read_data_cfg(datacfg);
srand(2222222);
@ -862,7 +747,7 @@ void threat_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_i
image in = get_image_from_stream(cap);
if(!in.data) break;
image in_s = resize_image(in, net.w, net.h);
image in_s = resize_image(in, net->w, net->h);
image out = in;
int x1 = out.w / 20;
@ -956,11 +841,8 @@ void gun_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_inde
int bad_cats[] = {218, 539, 540, 1213, 1501, 1742, 1911, 2415, 4348, 19223, 368, 369, 370, 1133, 1200, 1306, 2122, 2301, 2537, 2823, 3179, 3596, 3639, 4489, 5107, 5140, 5289, 6240, 6631, 6762, 7048, 7171, 7969, 7984, 7989, 8824, 8927, 9915, 10270, 10448, 13401, 15205, 18358, 18894, 18895, 19249, 19697};
printf("Classifier Demo\n");
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
list *options = read_data_cfg(datacfg);
srand(2222222);
@ -990,7 +872,7 @@ void gun_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_inde
gettimeofday(&tval_before, NULL);
image in = get_image_from_stream(cap);
image in_s = resize_image(in, net.w, net.h);
image in_s = resize_image(in, net->w, net->h);
show_image(in, "Threat Detection");
float *predictions = network_predict(net, in_s.data);
@ -1033,11 +915,8 @@ void demo_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_ind
{
#ifdef OPENCV
printf("Classifier Demo\n");
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
list *options = read_data_cfg(datacfg);
srand(2222222);
@ -1067,11 +946,11 @@ void demo_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_ind
gettimeofday(&tval_before, NULL);
image in = get_image_from_stream(cap);
image in_s = resize_image(in, net.w, net.h);
image in_s = resize_image(in, net->w, net->h);
show_image(in, "Classifier");
float *predictions = network_predict(net, in_s.data);
if(net.hierarchy) hierarchy_predictions(predictions, net.outputs, net.hierarchy, 1, 1);
if(net->hierarchy) hierarchy_predictions(predictions, net->outputs, net->hierarchy, 1, 1);
top_predictions(net, top, indexes);
printf("\033[2J");

70
examples/coco.c
View File

@ -17,17 +17,14 @@ void train_coco(char *cfgfile, char *weightfile)
char *base = basecfg(cfgfile);
printf("%s\n", base);
float avg_loss = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net.batch*net.subdivisions;
int i = *net.seen/imgs;
network *net = load_network(cfgfile, weightfile, 0);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = net->batch*net->subdivisions;
int i = *net->seen/imgs;
data train, buffer;
layer l = net.layers[net.n - 1];
layer l = net->layers[net->n - 1];
int side = l.side;
int classes = l.classes;
@ -38,8 +35,8 @@ void train_coco(char *cfgfile, char *weightfile)
char **paths = (char **)list_to_array(plist);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.n = imgs;
args.m = plist->size;
@ -49,15 +46,15 @@ void train_coco(char *cfgfile, char *weightfile)
args.d = &buffer;
args.type = REGION_DATA;
args.angle = net.angle;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.angle = net->angle;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
//while(i*imgs < N*120){
while(get_current_batch(net) < net.max_batches){
while(get_current_batch(net) < net->max_batches){
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -67,7 +64,7 @@ void train_coco(char *cfgfile, char *weightfile)
printf("Loaded: %lf seconds\n", sec(clock()-time));
/*
image im = float_to_image(net.w, net.h, 3, train.X.vals[113]);
image im = float_to_image(net->w, net->h, 3, train.X.vals[113]);
image copy = copy_image(im);
draw_coco(copy, train.y.vals[113], 7, "truth");
cvWaitKey(0);
@ -128,14 +125,11 @@ int get_coco_image_id(char *filename)
return atoi(p+1);
}
void validate_coco(char *cfgfile, char *weightfile)
void validate_coco(char *cfg, char *weights)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
srand(time(0));
char *base = "results/";
@ -144,7 +138,7 @@ void validate_coco(char *cfgfile, char *weightfile)
//list *plist = get_paths("/home/pjreddie/data/voc/test/2007_test.txt");
char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
int classes = l.classes;
int side = l.side;
@ -174,8 +168,8 @@ void validate_coco(char *cfgfile, char *weightfile)
pthread_t *thr = calloc(nthreads, sizeof(pthread_t));
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.type = IMAGE_DATA;
for(t = 0; t < nthreads; ++t){
@ -221,19 +215,16 @@ void validate_coco(char *cfgfile, char *weightfile)
void validate_coco_recall(char *cfgfile, char *weightfile)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
srand(time(0));
char *base = "results/comp4_det_test_";
list *plist = get_paths("/home/pjreddie/data/voc/test/2007_test.txt");
char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
int classes = l.classes;
int side = l.side;
@ -264,7 +255,7 @@ void validate_coco_recall(char *cfgfile, char *weightfile)
for(i = 0; i < m; ++i){
char *path = paths[i];
image orig = load_image_color(path, 0, 0);
image sized = resize_image(orig, net.w, net.h);
image sized = resize_image(orig, net->w, net->h);
char *id = basecfg(path);
network_predict(net, sized.data);
get_detection_boxes(l, 1, 1, thresh, probs, boxes, 1);
@ -309,12 +300,9 @@ void validate_coco_recall(char *cfgfile, char *weightfile)
void test_coco(char *cfgfile, char *weightfile, char *filename, float thresh)
{
image **alphabet = load_alphabet();
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
layer l = net.layers[net.n-1];
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
layer l = net->layers[net->n-1];
set_batch_network(net, 1);
srand(2222222);
float nms = .4;
clock_t time;
@ -335,7 +323,7 @@ void test_coco(char *cfgfile, char *weightfile, char *filename, float thresh)
strtok(input, "\n");
}
image im = load_image_color(input,0,0);
image sized = resize_image(im, net.w, net.h);
image sized = resize_image(im, net->w, net->h);
float *X = sized.data;
time=clock();
network_predict(net, X);

168
examples/darknet.c
View File

@ -6,20 +6,15 @@
extern void predict_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filename, int top);
extern void test_detector(char *datacfg, char *cfgfile, char *weightfile, char *filename, float thresh, float hier_thresh, char *outfile, int fullscreen);
extern void run_voxel(int argc, char **argv);
extern void run_yolo(int argc, char **argv);
extern void run_detector(int argc, char **argv);
extern void run_coco(int argc, char **argv);
extern void run_writing(int argc, char **argv);
extern void run_captcha(int argc, char **argv);
extern void run_nightmare(int argc, char **argv);
extern void run_dice(int argc, char **argv);
extern void run_compare(int argc, char **argv);
extern void run_classifier(int argc, char **argv);
extern void run_regressor(int argc, char **argv);
extern void run_segmenter(int argc, char **argv);
extern void run_char_rnn(int argc, char **argv);
extern void run_vid_rnn(int argc, char **argv);
extern void run_tag(int argc, char **argv);
extern void run_cifar(int argc, char **argv);
extern void run_go(int argc, char **argv);
@ -32,20 +27,20 @@ void average(int argc, char *argv[])
char *cfgfile = argv[2];
char *outfile = argv[3];
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
network sum = parse_network_cfg(cfgfile);
network *net = parse_network_cfg(cfgfile);
network *sum = parse_network_cfg(cfgfile);
char *weightfile = argv[4];
load_weights(&sum, weightfile);
load_weights(sum, weightfile);
int i, j;
int n = argc - 5;
for(i = 0; i < n; ++i){
weightfile = argv[i+5];
load_weights(&net, weightfile);
for(j = 0; j < net.n; ++j){
layer l = net.layers[j];
layer out = sum.layers[j];
load_weights(net, weightfile);
for(j = 0; j < net->n; ++j){
layer l = net->layers[j];
layer out = sum->layers[j];
if(l.type == CONVOLUTIONAL){
int num = l.n*l.c*l.size*l.size;
axpy_cpu(l.n, 1, l.biases, 1, out.biases, 1);
@ -63,8 +58,8 @@ void average(int argc, char *argv[])
}
}
n = n+1;
for(j = 0; j < net.n; ++j){
layer l = sum.layers[j];
for(j = 0; j < net->n; ++j){
layer l = sum->layers[j];
if(l.type == CONVOLUTIONAL){
int num = l.n*l.c*l.size*l.size;
scal_cpu(l.n, 1./n, l.biases, 1);
@ -83,12 +78,12 @@ void average(int argc, char *argv[])
save_weights(sum, outfile);
}
long numops(network net)
long numops(network *net)
{
int i;
long ops = 0;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
for(i = 0; i < net->n; ++i){
layer l = net->layers[i];
if(l.type == CONVOLUTIONAL){
ops += 2l * l.n * l.size*l.size*l.c/l.groups * l.out_h*l.out_w;
} else if(l.type == CONNECTED){
@ -121,11 +116,11 @@ long numops(network net)
void speed(char *cfgfile, int tics)
{
if (tics == 0) tics = 1000;
network net = parse_network_cfg(cfgfile);
set_batch_network(&net, 1);
network *net = parse_network_cfg(cfgfile);
set_batch_network(net, 1);
int i;
double time=what_time_is_it_now();
image im = make_image(net.w, net.h, net.c*net.batch);
image im = make_image(net->w, net->h, net->c*net->batch);
for(i = 0; i < tics; ++i){
network_predict(net, im.data);
}
@ -141,7 +136,7 @@ void speed(char *cfgfile, int tics)
void operations(char *cfgfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
network *net = parse_network_cfg(cfgfile);
long ops = numops(net);
printf("Floating Point Operations: %ld\n", ops);
printf("Floating Point Operations: %.2f Bn\n", (float)ops/1000000000.);
@ -150,63 +145,56 @@ void operations(char *cfgfile)
void oneoff(char *cfgfile, char *weightfile, char *outfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
int oldn = net.layers[net.n - 2].n;
int c = net.layers[net.n - 2].c;
scal_cpu(oldn*c, .1, net.layers[net.n - 2].weights, 1);
scal_cpu(oldn, 0, net.layers[net.n - 2].biases, 1);
net.layers[net.n - 2].n = 11921;
net.layers[net.n - 2].biases += 5;
net.layers[net.n - 2].weights += 5*c;
network *net = parse_network_cfg(cfgfile);
int oldn = net->layers[net->n - 2].n;
int c = net->layers[net->n - 2].c;
scal_cpu(oldn*c, .1, net->layers[net->n - 2].weights, 1);
scal_cpu(oldn, 0, net->layers[net->n - 2].biases, 1);
net->layers[net->n - 2].n = 11921;
net->layers[net->n - 2].biases += 5;
net->layers[net->n - 2].weights += 5*c;
if(weightfile){
load_weights(&net, weightfile);
load_weights(net, weightfile);
}
net.layers[net.n - 2].biases -= 5;
net.layers[net.n - 2].weights -= 5*c;
net.layers[net.n - 2].n = oldn;
net->layers[net->n - 2].biases -= 5;
net->layers[net->n - 2].weights -= 5*c;
net->layers[net->n - 2].n = oldn;
printf("%d\n", oldn);
layer l = net.layers[net.n - 2];
layer l = net->layers[net->n - 2];
copy_cpu(l.n/3, l.biases, 1, l.biases + l.n/3, 1);
copy_cpu(l.n/3, l.biases, 1, l.biases + 2*l.n/3, 1);
copy_cpu(l.n/3*l.c, l.weights, 1, l.weights + l.n/3*l.c, 1);
copy_cpu(l.n/3*l.c, l.weights, 1, l.weights + 2*l.n/3*l.c, 1);
*net.seen = 0;
*net->seen = 0;
save_weights(net, outfile);
}
void oneoff2(char *cfgfile, char *weightfile, char *outfile, int l)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
network *net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights_upto(&net, weightfile, 0, net.n);
load_weights_upto(&net, weightfile, l, net.n);
load_weights_upto(net, weightfile, 0, net->n);
load_weights_upto(net, weightfile, l, net->n);
}
*net.seen = 0;
save_weights_upto(net, outfile, net.n);
*net->seen = 0;
save_weights_upto(net, outfile, net->n);
}
void partial(char *cfgfile, char *weightfile, char *outfile, int max)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights_upto(&net, weightfile, 0, max);
}
*net.seen = 0;
network *net = load_network(cfgfile, weightfile, 1);
save_weights_upto(net, outfile, max);
}
void rescale_net(char *cfgfile, char *weightfile, char *outfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
int i;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
for(i = 0; i < net->n; ++i){
layer l = net->layers[i];
if(l.type == CONVOLUTIONAL){
rescale_weights(l, 2, -.5);
break;
@ -218,13 +206,10 @@ void rescale_net(char *cfgfile, char *weightfile, char *outfile)
void rgbgr_net(char *cfgfile, char *weightfile, char *outfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
int i;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
for(i = 0; i < net->n; ++i){
layer l = net->layers[i];
if(l.type == CONVOLUTIONAL){
rgbgr_weights(l);
break;
@ -236,13 +221,10 @@ void rgbgr_net(char *cfgfile, char *weightfile, char *outfile)
void reset_normalize_net(char *cfgfile, char *weightfile, char *outfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
if (weightfile) {
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
int i;
for (i = 0; i < net.n; ++i) {
layer l = net.layers[i];
for (i = 0; i < net->n; ++i) {
layer l = net->layers[i];
if (l.type == CONVOLUTIONAL && l.batch_normalize) {
denormalize_convolutional_layer(l);
}
@ -277,18 +259,15 @@ layer normalize_layer(layer l, int n)
void normalize_net(char *cfgfile, char *weightfile, char *outfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
int i;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
for(i = 0; i < net->n; ++i){
layer l = net->layers[i];
if(l.type == CONVOLUTIONAL && !l.batch_normalize){
net.layers[i] = normalize_layer(l, l.n);
net->layers[i] = normalize_layer(l, l.n);
}
if (l.type == CONNECTED && !l.batch_normalize) {
net.layers[i] = normalize_layer(l, l.outputs);
net->layers[i] = normalize_layer(l, l.outputs);
}
if (l.type == GRU && l.batch_normalize) {
*l.input_z_layer = normalize_layer(*l.input_z_layer, l.input_z_layer->outputs);
@ -297,7 +276,7 @@ void normalize_net(char *cfgfile, char *weightfile, char *outfile)
*l.state_z_layer = normalize_layer(*l.state_z_layer, l.state_z_layer->outputs);
*l.state_r_layer = normalize_layer(*l.state_r_layer, l.state_r_layer->outputs);
*l.state_h_layer = normalize_layer(*l.state_h_layer, l.state_h_layer->outputs);
net.layers[i].batch_normalize=1;
net->layers[i].batch_normalize=1;
}
}
save_weights(net, outfile);
@ -306,13 +285,10 @@ void normalize_net(char *cfgfile, char *weightfile, char *outfile)
void statistics_net(char *cfgfile, char *weightfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
if (weightfile) {
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
int i;
for (i = 0; i < net.n; ++i) {
layer l = net.layers[i];
for (i = 0; i < net->n; ++i) {
layer l = net->layers[i];
if (l.type == CONNECTED && l.batch_normalize) {
printf("Connected Layer %d\n", i);
statistics_connected_layer(l);
@ -339,20 +315,17 @@ void statistics_net(char *cfgfile, char *weightfile)
void denormalize_net(char *cfgfile, char *weightfile, char *outfile)
{
gpu_index = -1;
network net = parse_network_cfg(cfgfile);
if (weightfile) {
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
int i;
for (i = 0; i < net.n; ++i) {
layer l = net.layers[i];
for (i = 0; i < net->n; ++i) {
layer l = net->layers[i];
if ((l.type == DECONVOLUTIONAL || l.type == CONVOLUTIONAL) && l.batch_normalize) {
denormalize_convolutional_layer(l);
net.layers[i].batch_normalize=0;
net->layers[i].batch_normalize=0;
}
if (l.type == CONNECTED && l.batch_normalize) {
denormalize_connected_layer(l);
net.layers[i].batch_normalize=0;
net->layers[i].batch_normalize=0;
}
if (l.type == GRU && l.batch_normalize) {
denormalize_connected_layer(*l.input_z_layer);
@ -367,7 +340,7 @@ void denormalize_net(char *cfgfile, char *weightfile, char *outfile)
l.state_z_layer->batch_normalize = 0;
l.state_r_layer->batch_normalize = 0;
l.state_h_layer->batch_normalize = 0;
net.layers[i].batch_normalize=0;
net->layers[i].batch_normalize=0;
}
}
save_weights(net, outfile);
@ -375,9 +348,9 @@ void denormalize_net(char *cfgfile, char *weightfile, char *outfile)
void mkimg(char *cfgfile, char *weightfile, int h, int w, int num, char *prefix)
{
network net = load_network(cfgfile, weightfile, 0);
image *ims = get_weights(net.layers[0]);
int n = net.layers[0].n;
network *net = load_network(cfgfile, weightfile, 0);
image *ims = get_weights(net->layers[0]);
int n = net->layers[0].n;
int z;
for(z = 0; z < num; ++z){
image im = make_image(h, w, 3);
@ -401,10 +374,7 @@ void mkimg(char *cfgfile, char *weightfile, int h, int w, int num, char *prefix)
void visualize(char *cfgfile, char *weightfile)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, 0);
visualize_network(net);
#ifdef OPENCV
cvWaitKey(0);
@ -437,8 +407,6 @@ int main(int argc, char **argv)
average(argc, argv);
} else if (0 == strcmp(argv[1], "yolo")){
run_yolo(argc, argv);
} else if (0 == strcmp(argv[1], "voxel")){
run_voxel(argc, argv);
} else if (0 == strcmp(argv[1], "super")){
run_super(argc, argv);
} else if (0 == strcmp(argv[1], "lsd")){
@ -457,8 +425,6 @@ int main(int argc, char **argv)
run_go(argc, argv);
} else if (0 == strcmp(argv[1], "rnn")){
run_char_rnn(argc, argv);
} else if (0 == strcmp(argv[1], "vid")){
run_vid_rnn(argc, argv);
} else if (0 == strcmp(argv[1], "coco")){
run_coco(argc, argv);
} else if (0 == strcmp(argv[1], "classify")){
@ -473,12 +439,6 @@ int main(int argc, char **argv)
run_art(argc, argv);
} else if (0 == strcmp(argv[1], "tag")){
run_tag(argc, argv);
} else if (0 == strcmp(argv[1], "compare")){
run_compare(argc, argv);
} else if (0 == strcmp(argv[1], "dice")){
run_dice(argc, argv);
} else if (0 == strcmp(argv[1], "writing")){
run_writing(argc, argv);
} else if (0 == strcmp(argv[1], "3d")){
composite_3d(argv[2], argv[3], argv[4], (argc > 5) ? atof(argv[5]) : 0);
} else if (0 == strcmp(argv[1], "test")){

96
examples/detector.c
View File

@ -12,7 +12,7 @@ void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, i
char *base = basecfg(cfgfile);
printf("%s\n", base);
float avg_loss = -1;
network *nets = calloc(ngpus, sizeof(network));
network **nets = calloc(ngpus, sizeof(network));
srand(time(0));
int seed = rand();
@ -23,16 +23,16 @@ void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, i
cuda_set_device(gpus[i]);
#endif
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i].learning_rate *= ngpus;
nets[i]->learning_rate *= ngpus;
}
srand(time(0));
network net = nets[0];
network *net = nets[0];
int imgs = net.batch * net.subdivisions * ngpus;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net->batch * net->subdivisions * ngpus;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
data train, buffer;
layer l = net.layers[net.n - 1];
layer l = net->layers[net->n - 1];
int classes = l.classes;
float jitter = l.jitter;
@ -58,11 +58,11 @@ void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, i
double time;
int count = 0;
//while(i*imgs < N*120){
while(get_current_batch(net) < net.max_batches){
while(get_current_batch(net) < net->max_batches){
if(l.random && count++%10 == 0){
printf("Resizing\n");
int dim = (rand() % 10 + 10) * 32;
if (get_current_batch(net)+200 > net.max_batches) dim = 608;
if (get_current_batch(net)+200 > net->max_batches) dim = 608;
//int dim = (rand() % 4 + 16) * 32;
printf("%d\n", dim);
args.w = dim;
@ -74,7 +74,7 @@ void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, i
load_thread = load_data(args);
for(i = 0; i < ngpus; ++i){
resize_network(nets + i, dim, dim);
resize_network(nets[i], dim, dim);
}
net = nets[0];
}
@ -94,7 +94,7 @@ void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, i
/*
int zz;
for(zz = 0; zz < train.X.cols; ++zz){
image im = float_to_image(net.w, net.h, 3, train.X.vals[zz]);
image im = float_to_image(net->w, net->h, 3, train.X.vals[zz]);
int k;
for(k = 0; k < l.max_boxes; ++k){
box b = float_to_box(train.y.vals[zz] + k*5, 1);
@ -239,18 +239,15 @@ void validate_detector_flip(char *datacfg, char *cfgfile, char *weightfile, char
int *map = 0;
if (mapf) map = read_map(mapf);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 2);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 2);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
srand(time(0));
list *plist = get_paths(valid_images);
char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
int classes = l.classes;
char buff[1024];
@ -299,11 +296,11 @@ void validate_detector_flip(char *datacfg, char *cfgfile, char *weightfile, char
image *buf_resized = calloc(nthreads, sizeof(image));
pthread_t *thr = calloc(nthreads, sizeof(pthread_t));
image input = make_image(net.w, net.h, net.c*2);
image input = make_image(net->w, net->h, net->c*2);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
//args.type = IMAGE_DATA;
args.type = LETTERBOX_DATA;
@ -330,14 +327,14 @@ void validate_detector_flip(char *datacfg, char *cfgfile, char *weightfile, char
for(t = 0; t < nthreads && i+t-nthreads < m; ++t){
char *path = paths[i+t-nthreads];
char *id = basecfg(path);
copy_cpu(net.w*net.h*net.c, val_resized[t].data, 1, input.data, 1);
copy_cpu(net->w*net->h*net->c, val_resized[t].data, 1, input.data, 1);
flip_image(val_resized[t]);
copy_cpu(net.w*net.h*net.c, val_resized[t].data, 1, input.data + net.w*net.h*net.c, 1);
copy_cpu(net->w*net->h*net->c, val_resized[t].data, 1, input.data + net->w*net->h*net->c, 1);
network_predict(net, input.data);
int w = val[t].w;
int h = val[t].h;
get_region_boxes(l, w, h, net.w, net.h, thresh, probs, boxes, 0, 0, map, .5, 0);
get_region_boxes(l, w, h, net->w, net->h, thresh, probs, boxes, 0, 0, map, .5, 0);
if (nms) do_nms_sort(boxes, probs, l.w*l.h*l.n, classes, nms);
if (coco){
print_cocos(fp, path, boxes, probs, l.w*l.h*l.n, classes, w, h);
@ -375,18 +372,15 @@ void validate_detector(char *datacfg, char *cfgfile, char *weightfile, char *out
int *map = 0;
if (mapf) map = read_map(mapf);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
srand(time(0));
list *plist = get_paths(valid_images);
char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
int classes = l.classes;
char buff[1024];
@ -436,8 +430,8 @@ void validate_detector(char *datacfg, char *cfgfile, char *weightfile, char *out
pthread_t *thr = calloc(nthreads, sizeof(pthread_t));
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
//args.type = IMAGE_DATA;
args.type = LETTERBOX_DATA;
@ -468,7 +462,7 @@ void validate_detector(char *datacfg, char *cfgfile, char *weightfile, char *out
network_predict(net, X);
int w = val[t].w;
int h = val[t].h;
get_region_boxes(l, w, h, net.w, net.h, thresh, probs, boxes, 0, 0, map, .5, 0);
get_region_boxes(l, w, h, net->w, net->h, thresh, probs, boxes, 0, 0, map, .5, 0);
if (nms) do_nms_sort(boxes, probs, l.w*l.h*l.n, classes, nms);
if (coco){
print_cocos(fp, path, boxes, probs, l.w*l.h*l.n, classes, w, h);
@ -495,18 +489,15 @@ void validate_detector(char *datacfg, char *cfgfile, char *weightfile, char *out
void validate_detector_recall(char *cfgfile, char *weightfile)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
srand(time(0));
list *plist = get_paths("data/coco_val_5k.list");
char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
int classes = l.classes;
int j, k;
@ -529,10 +520,10 @@ void validate_detector_recall(char *cfgfile, char *weightfile)
for(i = 0; i < m; ++i){
char *path = paths[i];
image orig = load_image_color(path, 0, 0);
image sized = resize_image(orig, net.w, net.h);
image sized = resize_image(orig, net->w, net->h);
char *id = basecfg(path);
network_predict(net, sized.data);
get_region_boxes(l, sized.w, sized.h, net.w, net.h, thresh, probs, boxes, 0, 1, 0, .5, 1);
get_region_boxes(l, sized.w, sized.h, net->w, net->h, thresh, probs, boxes, 0, 1, 0, .5, 1);
if (nms) do_nms(boxes, probs, l.w*l.h*l.n, 1, nms);
char labelpath[4096];
@ -578,11 +569,8 @@ void test_detector(char *datacfg, char *cfgfile, char *weightfile, char *filenam
char **names = get_labels(name_list);
image **alphabet = load_alphabet();
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
double time;
char buff[256];
@ -600,12 +588,12 @@ void test_detector(char *datacfg, char *cfgfile, char *weightfile, char *filenam
strtok(input, "\n");
}
image im = load_image_color(input,0,0);
image sized = letterbox_image(im, net.w, net.h);
//image sized = resize_image(im, net.w, net.h);
//image sized2 = resize_max(im, net.w);
//image sized = crop_image(sized2, -((net.w - sized2.w)/2), -((net.h - sized2.h)/2), net.w, net.h);
//resize_network(&net, sized.w, sized.h);
layer l = net.layers[net.n-1];
image sized = letterbox_image(im, net->w, net->h);
//image sized = resize_image(im, net->w, net->h);
//image sized2 = resize_max(im, net->w);
//image sized = crop_image(sized2, -((net->w - sized2.w)/2), -((net->h - sized2.h)/2), net->w, net->h);
//resize_network(net, sized.w, sized.h);
layer l = net->layers[net->n-1];
box *boxes = calloc(l.w*l.h*l.n, sizeof(box));
float **probs = calloc(l.w*l.h*l.n, sizeof(float *));
@ -620,7 +608,7 @@ void test_detector(char *datacfg, char *cfgfile, char *weightfile, char *filenam
time=what_time_is_it_now();
network_predict(net, X);
printf("%s: Predicted in %f seconds.\n", input, what_time_is_it_now()-time);
get_region_boxes(l, im.w, im.h, net.w, net.h, thresh, probs, boxes, masks, 0, 0, hier_thresh, 1);
get_region_boxes(l, im.w, im.h, net->w, net->h, thresh, probs, boxes, masks, 0, 0, hier_thresh, 1);
if (nms) do_nms_sort(boxes, probs, l.w*l.h*l.n, l.classes, nms);
//else if (nms) do_nms_sort(boxes, probs, l.w*l.h*l.n, l.classes, nms);
draw_detections(im, l.w*l.h*l.n, thresh, boxes, probs, masks, names, alphabet, l.classes);

88
examples/go.c
View File

@ -124,7 +124,7 @@ void train_go(char *cfgfile, char *weightfile, char *filename, int *gpus, int ng
char *base = basecfg(cfgfile);
printf("%s\n", base);
printf("%d\n", ngpus);
network *nets = calloc(ngpus, sizeof(network));
network **nets = calloc(ngpus, sizeof(network*));
srand(time(0));
int seed = rand();
@ -134,10 +134,10 @@ void train_go(char *cfgfile, char *weightfile, char *filename, int *gpus, int ng
cuda_set_device(gpus[i]);
#endif
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i].learning_rate *= ngpus;
nets[i]->learning_rate *= ngpus;
}
network net = nets[0];
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = nets[0];
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
char *backup_directory = "/home/pjreddie/backup/";
@ -147,11 +147,11 @@ void train_go(char *cfgfile, char *weightfile, char *filename, int *gpus, int ng
int N = m.n;
printf("Moves: %d\n", N);
int epoch = (*net.seen)/N;
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
int epoch = (*net->seen)/N;
while(get_current_batch(net) < net->max_batches || net->max_batches == 0){
clock_t time=clock();
data train = random_go_moves(m, net.batch*net.subdivisions*ngpus);
data train = random_go_moves(m, net->batch*net->subdivisions*ngpus);
printf("Loaded: %lf seconds\n", sec(clock()-time));
time=clock();
@ -169,9 +169,9 @@ void train_go(char *cfgfile, char *weightfile, char *filename, int *gpus, int ng
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.95 + loss*.05;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net->seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net->seen);
if(*net->seen/N > epoch){
epoch = *net->seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory,base, epoch);
save_weights(net, buff);
@ -281,7 +281,7 @@ void flip_board(float *board)
}
}
void predict_move(network net, float *board, float *move, int multi)
void predict_move(network *net, float *board, float *move, int multi)
{
float *output = network_predict(net, board);
copy_cpu(19*19+1, output, 1, move, 1);
@ -370,7 +370,7 @@ int legal_go(float *b, char *ko, int p, int r, int c)
return 1;
}
int generate_move(network net, int player, float *board, int multi, float thresh, float temp, char *ko, int print)
int generate_move(network *net, int player, float *board, int multi, float thresh, float temp, char *ko, int print)
{
int i, j;
int empty = 1;
@ -383,7 +383,7 @@ int generate_move(network net, int player, float *board, int multi, float thresh
if(empty) {
return 72;
}
for(i = 0; i < net.n; ++i) net.layers[i].temperature = temp;
for(i = 0; i < net->n; ++i) net->layers[i].temperature = temp;
float move[362];
if (player < 0) flip_board(board);
@ -439,12 +439,9 @@ void valid_go(char *cfgfile, char *weightfile, int multi, char *filename)
srand(time(0));
char *base = basecfg(cfgfile);
printf("%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
float *board = calloc(19*19, sizeof(float));
float *move = calloc(19*19+1, sizeof(float));
@ -486,12 +483,9 @@ int print_game(float *board, FILE *fp)
void engine_go(char *filename, char *weightfile, int multi)
{
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
set_batch_network(net, 1);
srand(time(0));
set_batch_network(&net, 1);
float *board = calloc(19*19, sizeof(float));
char *one = calloc(91, sizeof(char));
char *two = calloc(91, sizeof(char));
@ -679,12 +673,9 @@ void engine_go(char *filename, char *weightfile, int multi)
void test_go(char *cfg, char *weights, int multi)
{
network net = parse_network_cfg(cfg);
if(weights){
load_weights(&net, weights);
}
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
srand(time(0));
set_batch_network(&net, 1);
float *board = calloc(19*19, sizeof(float));
float *move = calloc(19*19+1, sizeof(float));
int color = 1;
@ -785,23 +776,24 @@ float score_game(float *board)
void self_go(char *filename, char *weightfile, char *f2, char *w2, int multi)
{
network net = parse_network_cfg(filename);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(filename, weightfile, 0);
set_batch_network(net, 1);
network net2 = net;
if(f2){
network *net2;
if (f2) {
net2 = parse_network_cfg(f2);
if(w2){
load_weights(&net2, w2);
load_weights(net2, w2);
}
} else {
net2 = calloc(1, sizeof(network));
*net2 = *net;
}
srand(time(0));
char boards[600][93];
int count = 0;
set_batch_network(&net, 1);
set_batch_network(&net2, 1);
set_batch_network(net, 1);
set_batch_network(net2, 1);
float *board = calloc(19*19, sizeof(float));
char *one = calloc(91, sizeof(char));
char *two = calloc(91, sizeof(char));
@ -819,15 +811,15 @@ void self_go(char *filename, char *weightfile, char *f2, char *w2, int multi)
fprintf(stderr, "Total: %d, Player 1: %f, Player 2: %f\n", total, (float)p1/total, (float)p2/total);
sleep(1);
/*
int i = (score > 0)? 0 : 1;
int j;
for(; i < count; i += 2){
for(j = 0; j < 93; ++j){
printf("%c", boards[i][j]);
}
printf("\n");
}
*/
int i = (score > 0)? 0 : 1;
int j;
for(; i < count; i += 2){
for(j = 0; j < 93; ++j){
printf("%c", boards[i][j]);
}
printf("\n");
}
*/
memset(board, 0, 19*19*sizeof(float));
player = 1;
done = 0;
@ -837,7 +829,7 @@ void self_go(char *filename, char *weightfile, char *f2, char *w2, int multi)
}
print_board(stderr, board, 1, 0);
//sleep(1);
network use = ((total%2==0) == (player==1)) ? net : net2;
network *use = ((total%2==0) == (player==1)) ? net : net2;
int index = generate_move(use, player, board, multi, .4, 1, two, 0);
if(index < 0){
done = 1;

368
examples/lsd.c
View File

@ -16,9 +16,9 @@ void train_lsd3(char *fcfg, char *fweight, char *gcfg, char *gweight, char *acfg
char *gbase = basecfg(gcfg);
char *abase = basecfg(acfg);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", gnet.learning_rate, gnet.momentum, gnet.decay);
int imgs = gnet.batch*gnet.subdivisions;
int i = *gnet.seen/imgs;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", gnet->learning_rate, gnet->momentum, gnet->decay);
int imgs = gnet->batch*gnet->subdivisions;
int i = *gnet->seen/imgs;
data train, tbuffer;
data style, sbuffer;
@ -55,27 +55,27 @@ void train_lsd3(char *fcfg, char *fweight, char *gcfg, char *gweight, char *acfg
float aloss_avg = -1;
float floss_avg = -1;
fnet.train=1;
int x_size = fnet.inputs*fnet.batch;
int y_size = fnet.truths*fnet.batch;
fnet->train=1;
int x_size = fnet->inputs*fnet->batch;
int y_size = fnet->truths*fnet->batch;
float *X = calloc(x_size, sizeof(float));
float *y = calloc(y_size, sizeof(float));
int ax_size = anet.inputs*anet.batch;
int ay_size = anet.truths*anet.batch;
fill_gpu(ay_size, .9, anet.truth_gpu, 1);
anet.delta_gpu = cuda_make_array(0, ax_size);
anet.train = 1;
int ax_size = anet->inputs*anet->batch;
int ay_size = anet->truths*anet->batch;
fill_gpu(ay_size, .9, anet->truth_gpu, 1);
anet->delta_gpu = cuda_make_array(0, ax_size);
anet->train = 1;
int gx_size = gnet.inputs*gnet.batch;
int gy_size = gnet.truths*gnet.batch;
int gx_size = gnet->inputs*gnet->batch;
int gy_size = gnet->truths*gnet->batch;
gstate.input = cuda_make_array(0, gx_size);
gstate.truth = 0;
gstate.delta = 0;
gstate.train = 1;
while (get_current_batch(gnet) < gnet.max_batches) {
while (get_current_batch(gnet) < gnet->max_batches) {
i += 1;
time=clock();
pthread_join(tload_thread, 0);
@ -92,20 +92,20 @@ void train_lsd3(char *fcfg, char *fweight, char *gcfg, char *gweight, char *acfg
int j, k;
float floss = 0;
for(j = 0; j < fnet.subdivisions; ++j){
layer imlayer = gnet.layers[gnet.n - 1];
get_next_batch(train, fnet.batch, j*fnet.batch, X, y);
for(j = 0; j < fnet->subdivisions; ++j){
layer imlayer = gnet->layers[gnet->n - 1];
get_next_batch(train, fnet->batch, j*fnet->batch, X, y);
cuda_push_array(fstate.input, X, x_size);
cuda_push_array(gstate.input, X, gx_size);
*gnet.seen += gnet.batch;
*gnet->seen += gnet->batch;
forward_network_gpu(fnet, fstate);
float *feats = fnet.layers[fnet.n - 2].output_gpu;
float *feats = fnet->layers[fnet->n - 2].output_gpu;
copy_gpu(y_size, feats, 1, fstate.truth, 1);
forward_network_gpu(gnet, gstate);
float *gen = gnet.layers[gnet.n-1].output_gpu;
float *gen = gnet->layers[gnet->n-1].output_gpu;
copy_gpu(x_size, gen, 1, fstate.input, 1);
fill_gpu(x_size, 0, fstate.delta, 1);
@ -135,11 +135,11 @@ void train_lsd3(char *fcfg, char *fweight, char *gcfg, char *gweight, char *acfg
backward_network_gpu(gnet, gstate);
floss += get_network_cost(fnet) /(fnet.subdivisions*fnet.batch);
floss += get_network_cost(fnet) /(fnet->subdivisions*fnet->batch);
cuda_pull_array(imlayer.output_gpu, imlayer.output, imlayer.outputs*imlayer.batch);
for(k = 0; k < gnet.batch; ++k){
int index = j*gnet.batch + k;
for(k = 0; k < gnet->batch; ++k){
int index = j*gnet->batch + k;
copy_cpu(imlayer.outputs, imlayer.output + k*imlayer.outputs, 1, generated.X.vals[index], 1);
generated.y.vals[index][0] = .1;
style.y.vals[index][0] = .9;
@ -148,7 +148,7 @@ void train_lsd3(char *fcfg, char *fweight, char *gcfg, char *gweight, char *acfg
*/
/*
image sim = float_to_image(anet.w, anet.h, anet.c, style.X.vals[j]);
image sim = float_to_image(anet->w, anet->h, anet->c, style.X.vals[j]);
show_image(sim, "style");
cvWaitKey(0);
*/
@ -208,16 +208,16 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
int i, j, k;
layer imlayer = {0};
for (i = 0; i < net.n; ++i) {
if (net.layers[i].out_c == 3) {
imlayer = net.layers[i];
for (i = 0; i < net->n; ++i) {
if (net->layers[i].out_c == 3) {
imlayer = net->layers[i];
break;
}
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net.batch*net.subdivisions;
i = *net.seen/imgs;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = net->batch*net->subdivisions;
i = *net->seen/imgs;
data train, buffer;
@ -226,21 +226,21 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
char **paths = (char **)list_to_array(plist);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.d = &buffer;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.size = net.w;
args.min = net->min_crop;
args.max = net->max_crop;
args.angle = net->angle;
args.aspect = net->aspect;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
args.size = net->w;
args.type = CLASSIFICATION_DATA;
args.classes = 1;
char *ls[1] = {"coco"};
@ -252,7 +252,7 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
network_state gstate = {0};
gstate.index = 0;
gstate.net = net;
int x_size = get_network_input_size(net)*net.batch;
int x_size = get_network_input_size(net)*net->batch;
int y_size = x_size;
gstate.input = cuda_make_array(0, x_size);
gstate.truth = cuda_make_array(0, y_size);
@ -265,7 +265,7 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
network_state astate = {0};
astate.index = 0;
astate.net = anet;
int ay_size = get_network_output_size(anet)*anet.batch;
int ay_size = get_network_output_size(anet)*anet->batch;
astate.input = 0;
astate.truth = 0;
astate.delta = 0;
@ -280,7 +280,7 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
//data generated = copy_data(train);
while (get_current_batch(net) < net.max_batches) {
while (get_current_batch(net) < net->max_batches) {
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -291,31 +291,31 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
data gray = copy_data(train);
for(j = 0; j < imgs; ++j){
image gim = float_to_image(net.w, net.h, net.c, gray.X.vals[j]);
image gim = float_to_image(net->w, net->h, net->c, gray.X.vals[j]);
grayscale_image_3c(gim);
train.y.vals[j][0] = .9;
image yim = float_to_image(net.w, net.h, net.c, train.X.vals[j]);
image yim = float_to_image(net->w, net->h, net->c, train.X.vals[j]);
//rgb_to_yuv(yim);
}
time=clock();
float gloss = 0;
for(j = 0; j < net.subdivisions; ++j){
get_next_batch(train, net.batch, j*net.batch, pixs, y);
get_next_batch(gray, net.batch, j*net.batch, graypixs, y);
for(j = 0; j < net->subdivisions; ++j){
get_next_batch(train, net->batch, j*net->batch, pixs, y);
get_next_batch(gray, net->batch, j*net->batch, graypixs, y);
cuda_push_array(gstate.input, graypixs, x_size);
cuda_push_array(gstate.truth, pixs, y_size);
*/
/*
image origi = float_to_image(net.w, net.h, 3, pixs);
image grayi = float_to_image(net.w, net.h, 3, graypixs);
image origi = float_to_image(net->w, net->h, 3, pixs);
image grayi = float_to_image(net->w, net->h, 3, graypixs);
show_image(grayi, "gray");
show_image(origi, "orig");
cvWaitKey(0);
*/
/*
*net.seen += net.batch;
*net->seen += net->batch;
forward_network_gpu(net, gstate);
fill_gpu(imlayer.outputs, 0, imerror, 1);
@ -325,22 +325,22 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
forward_network_gpu(anet, astate);
backward_network_gpu(anet, astate);
scal_gpu(imlayer.outputs, .1, net.layers[net.n-1].delta_gpu, 1);
scal_gpu(imlayer.outputs, .1, net->layers[net->n-1].delta_gpu, 1);
backward_network_gpu(net, gstate);
scal_gpu(imlayer.outputs, 1000, imerror, 1);
printf("realness %f\n", cuda_mag_array(imerror, imlayer.outputs));
printf("features %f\n", cuda_mag_array(net.layers[net.n-1].delta_gpu, imlayer.outputs));
printf("features %f\n", cuda_mag_array(net->layers[net->n-1].delta_gpu, imlayer.outputs));
axpy_gpu(imlayer.outputs, 1, imerror, 1, imlayer.delta_gpu, 1);
gloss += get_network_cost(net) /(net.subdivisions*net.batch);
gloss += get_network_cost(net) /(net->subdivisions*net->batch);
cuda_pull_array(imlayer.output_gpu, imlayer.output, imlayer.outputs*imlayer.batch);
for(k = 0; k < net.batch; ++k){
int index = j*net.batch + k;
for(k = 0; k < net->batch; ++k){
int index = j*net->batch + k;
copy_cpu(imlayer.outputs, imlayer.output + k*imlayer.outputs, 1, gray.X.vals[index], 1);
gray.y.vals[index][0] = .1;
}
@ -385,11 +385,8 @@ void train_pix2pix(char *cfg, char *weight, char *acfg, char *aweight, int clear
void test_dcgan(char *cfgfile, char *weightfile)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
clock_t time;
@ -397,8 +394,8 @@ void test_dcgan(char *cfgfile, char *weightfile)
char *input = buff;
int i, imlayer = 0;
for (i = 0; i < net.n; ++i) {
if (net.layers[i].out_c == 3) {
for (i = 0; i < net->n; ++i) {
if (net->layers[i].out_c == 3) {
imlayer = i;
printf("%d\n", i);
break;
@ -406,7 +403,7 @@ void test_dcgan(char *cfgfile, char *weightfile)
}
while(1){
image im = make_image(net.w, net.h, net.c);
image im = make_image(net->w, net->h, net->c);
int i;
for(i = 0; i < im.w*im.h*im.c; ++i){
im.data[i] = rand_normal();
@ -449,23 +446,23 @@ void train_dcgan(char *cfg, char *weight, char *acfg, char *aweight, int clear,
char *base = basecfg(cfg);
char *abase = basecfg(acfg);
printf("%s\n", base);
network gnet = load_network(cfg, weight, clear);
network anet = load_network(acfg, aweight, clear);
//float orig_rate = anet.learning_rate;
network *gnet = load_network(cfg, weight, clear);
network *anet = load_network(acfg, aweight, clear);
//float orig_rate = anet->learning_rate;
int start = 0;
int i, j, k;
layer imlayer = {0};
for (i = 0; i < gnet.n; ++i) {
if (gnet.layers[i].out_c == 3) {
imlayer = gnet.layers[i];
for (i = 0; i < gnet->n; ++i) {
if (gnet->layers[i].out_c == 3) {
imlayer = gnet->layers[i];
break;
}
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", gnet.learning_rate, gnet.momentum, gnet.decay);
int imgs = gnet.batch*gnet.subdivisions;
i = *gnet.seen/imgs;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", gnet->learning_rate, gnet->momentum, gnet->decay);
int imgs = gnet->batch*gnet->subdivisions;
i = *gnet->seen/imgs;
data train, buffer;
@ -487,20 +484,20 @@ void train_dcgan(char *cfg, char *weight, char *acfg, char *aweight, int clear,
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
gnet.train = 1;
anet.train = 1;
gnet->train = 1;
anet->train = 1;
int x_size = gnet.inputs*gnet.batch;
int y_size = gnet.truths*gnet.batch;
int x_size = gnet->inputs*gnet->batch;
int y_size = gnet->truths*gnet->batch;
float *imerror = cuda_make_array(0, y_size);
//int ay_size = anet.truths*anet.batch;
//int ay_size = anet->truths*anet->batch;
float aloss_avg = -1;
//data generated = copy_data(train);
while (get_current_batch(gnet) < gnet.max_batches) {
while (get_current_batch(gnet) < gnet->max_batches) {
start += 1;
i += 1;
time=clock();
@ -521,41 +518,41 @@ void train_dcgan(char *cfg, char *weight, char *acfg, char *aweight, int clear,
}
time=clock();
for(j = 0; j < gnet.subdivisions; ++j){
get_next_batch(train, gnet.batch, j*gnet.batch, gnet.truth, 0);
for(j = 0; j < gnet->subdivisions; ++j){
get_next_batch(train, gnet->batch, j*gnet->batch, gnet->truth, 0);
int z;
for(z = 0; z < x_size; ++z){
gnet.input[z] = rand_normal();
gnet->input[z] = rand_normal();
}
cuda_push_array(gnet.input_gpu, gnet.input, x_size);
cuda_push_array(gnet.truth_gpu, gnet.truth, y_size);
*gnet.seen += gnet.batch;
cuda_push_array(gnet->input_gpu, gnet->input, x_size);
cuda_push_array(gnet->truth_gpu, gnet->truth, y_size);
*gnet->seen += gnet->batch;
forward_network_gpu(gnet);
fill_gpu(imlayer.outputs*imlayer.batch, 0, imerror, 1);
fill_gpu(anet.truths*anet.batch, .95, anet.truth_gpu, 1);
copy_gpu(anet.inputs*anet.batch, imlayer.output_gpu, 1, anet.input_gpu, 1);
anet.delta_gpu = imerror;
fill_gpu(anet->truths*anet->batch, .95, anet->truth_gpu, 1);
copy_gpu(anet->inputs*anet->batch, imlayer.output_gpu, 1, anet->input_gpu, 1);
anet->delta_gpu = imerror;
forward_network_gpu(anet);
backward_network_gpu(anet);
float genaloss = *anet.cost / anet.batch;
float genaloss = *anet->cost / anet->batch;
printf("%f\n", genaloss);
scal_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1);
scal_gpu(imlayer.outputs*imlayer.batch, .00, gnet.layers[gnet.n-1].delta_gpu, 1);
scal_gpu(imlayer.outputs*imlayer.batch, .00, gnet->layers[gnet->n-1].delta_gpu, 1);
printf("realness %f\n", cuda_mag_array(imerror, imlayer.outputs*imlayer.batch));
printf("features %f\n", cuda_mag_array(gnet.layers[gnet.n-1].delta_gpu, imlayer.outputs*imlayer.batch));
printf("features %f\n", cuda_mag_array(gnet->layers[gnet->n-1].delta_gpu, imlayer.outputs*imlayer.batch));
axpy_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1, gnet.layers[gnet.n-1].delta_gpu, 1);
axpy_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1, gnet->layers[gnet->n-1].delta_gpu, 1);
backward_network_gpu(gnet);
for(k = 0; k < gnet.batch; ++k){
int index = j*gnet.batch + k;
copy_cpu(gnet.outputs, gnet.output + k*gnet.outputs, 1, gen.X.vals[index], 1);
for(k = 0; k < gnet->batch; ++k){
int index = j*gnet->batch + k;
copy_cpu(gnet->outputs, gnet->output + k*gnet->outputs, 1, gen.X.vals[index], 1);
}
}
harmless_update_network_gpu(anet);
@ -570,8 +567,8 @@ void train_dcgan(char *cfg, char *weight, char *acfg, char *aweight, int clear,
//scale_image(im2, .5);
#ifdef OPENCV
if(display){
image im = float_to_image(anet.w, anet.h, anet.c, gen.X.vals[0]);
image im2 = float_to_image(anet.w, anet.h, anet.c, train.X.vals[0]);
image im = float_to_image(anet->w, anet->h, anet->c, gen.X.vals[0]);
image im2 = float_to_image(anet->w, anet->h, anet->c, train.X.vals[0]);
show_image(im, "gen");
show_image(im2, "train");
cvWaitKey(50);
@ -580,9 +577,9 @@ void train_dcgan(char *cfg, char *weight, char *acfg, char *aweight, int clear,
/*
if(aloss < .1){
anet.learning_rate = 0;
anet->learning_rate = 0;
} else if (aloss > .3){
anet.learning_rate = orig_rate;
anet->learning_rate = orig_rate;
}
*/
@ -627,21 +624,21 @@ void train_colorizer(char *cfg, char *weight, char *acfg, char *aweight, int cle
char *base = basecfg(cfg);
char *abase = basecfg(acfg);
printf("%s\n", base);
network net = load_network(cfg, weight, clear);
network anet = load_network(acfg, aweight, clear);
network *net = load_network(cfg, weight, clear);
network *anet = load_network(acfg, aweight, clear);
int i, j, k;
layer imlayer = {0};
for (i = 0; i < net.n; ++i) {
if (net.layers[i].out_c == 3) {
imlayer = net.layers[i];
for (i = 0; i < net->n; ++i) {
if (net->layers[i].out_c == 3) {
imlayer = net->layers[i];
break;
}
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net.batch*net.subdivisions;
i = *net.seen/imgs;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = net->batch*net->subdivisions;
i = *net->seen/imgs;
data train, buffer;
@ -663,17 +660,17 @@ void train_colorizer(char *cfg, char *weight, char *acfg, char *aweight, int cle
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
int x_size = net.inputs*net.batch;
int x_size = net->inputs*net->batch;
//int y_size = x_size;
net.delta = 0;
net.train = 1;
net->delta = 0;
net->train = 1;
float *pixs = calloc(x_size, sizeof(float));
float *graypixs = calloc(x_size, sizeof(float));
//float *y = calloc(y_size, sizeof(float));
//int ay_size = anet.outputs*anet.batch;
anet.delta = 0;
anet.train = 1;
//int ay_size = anet->outputs*anet->batch;
anet->delta = 0;
anet->train = 1;
float *imerror = cuda_make_array(0, imlayer.outputs*imlayer.batch);
@ -682,7 +679,7 @@ void train_colorizer(char *cfg, char *weight, char *acfg, char *aweight, int cle
//data generated = copy_data(train);
while (get_current_batch(net) < net.max_batches) {
while (get_current_batch(net) < net->max_batches) {
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -693,7 +690,7 @@ void train_colorizer(char *cfg, char *weight, char *acfg, char *aweight, int cle
data gray = copy_data(train);
for(j = 0; j < imgs; ++j){
image gim = float_to_image(net.w, net.h, net.c, gray.X.vals[j]);
image gim = float_to_image(net->w, net->h, net->c, gray.X.vals[j]);
grayscale_image_3c(gim);
train.y.vals[j][0] = .95;
gray.y.vals[j][0] = .05;
@ -701,44 +698,44 @@ void train_colorizer(char *cfg, char *weight, char *acfg, char *aweight, int cle
time=clock();
float gloss = 0;
for(j = 0; j < net.subdivisions; ++j){
get_next_batch(train, net.batch, j*net.batch, pixs, 0);
get_next_batch(gray, net.batch, j*net.batch, graypixs, 0);
cuda_push_array(net.input_gpu, graypixs, net.inputs*net.batch);
cuda_push_array(net.truth_gpu, pixs, net.truths*net.batch);
for(j = 0; j < net->subdivisions; ++j){
get_next_batch(train, net->batch, j*net->batch, pixs, 0);
get_next_batch(gray, net->batch, j*net->batch, graypixs, 0);
cuda_push_array(net->input_gpu, graypixs, net->inputs*net->batch);
cuda_push_array(net->truth_gpu, pixs, net->truths*net->batch);
/*
image origi = float_to_image(net.w, net.h, 3, pixs);
image grayi = float_to_image(net.w, net.h, 3, graypixs);
image origi = float_to_image(net->w, net->h, 3, pixs);
image grayi = float_to_image(net->w, net->h, 3, graypixs);
show_image(grayi, "gray");
show_image(origi, "orig");
cvWaitKey(0);
*/
*net.seen += net.batch;
*net->seen += net->batch;
forward_network_gpu(net);
fill_gpu(imlayer.outputs*imlayer.batch, 0, imerror, 1);
copy_gpu(anet.inputs*anet.batch, imlayer.output_gpu, 1, anet.input_gpu, 1);
fill_gpu(anet.inputs*anet.batch, .95, anet.truth_gpu, 1);
anet.delta_gpu = imerror;
copy_gpu(anet->inputs*anet->batch, imlayer.output_gpu, 1, anet->input_gpu, 1);
fill_gpu(anet->inputs*anet->batch, .95, anet->truth_gpu, 1);
anet->delta_gpu = imerror;
forward_network_gpu(anet);
backward_network_gpu(anet);
scal_gpu(imlayer.outputs*imlayer.batch, 1./100., net.layers[net.n-1].delta_gpu, 1);
scal_gpu(imlayer.outputs*imlayer.batch, 1./100., net->layers[net->n-1].delta_gpu, 1);
scal_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1);
printf("realness %f\n", cuda_mag_array(imerror, imlayer.outputs*imlayer.batch));
printf("features %f\n", cuda_mag_array(net.layers[net.n-1].delta_gpu, imlayer.outputs*imlayer.batch));
printf("features %f\n", cuda_mag_array(net->layers[net->n-1].delta_gpu, imlayer.outputs*imlayer.batch));
axpy_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1, net.layers[net.n-1].delta_gpu, 1);
axpy_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1, net->layers[net->n-1].delta_gpu, 1);
backward_network_gpu(net);
gloss += *net.cost /(net.subdivisions*net.batch);
gloss += *net->cost /(net->subdivisions*net->batch);
for(k = 0; k < net.batch; ++k){
int index = j*net.batch + k;
for(k = 0; k < net->batch; ++k){
int index = j*net->batch + k;
copy_cpu(imlayer.outputs, imlayer.output + k*imlayer.outputs, 1, gray.X.vals[index], 1);
}
}
@ -752,8 +749,8 @@ void train_colorizer(char *cfg, char *weight, char *acfg, char *aweight, int cle
#ifdef OPENCV
if(display){
image im = float_to_image(anet.w, anet.h, anet.c, gray.X.vals[0]);
image im2 = float_to_image(anet.w, anet.h, anet.c, train.X.vals[0]);
image im = float_to_image(anet->w, anet->h, anet->c, gray.X.vals[0]);
image im2 = float_to_image(anet->w, anet->h, anet->c, train.X.vals[0]);
show_image(im, "gen");
show_image(im2, "train");
cvWaitKey(50);
@ -801,27 +798,27 @@ void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfi
if(weightfile){
load_weights(&net, weightfile);
}
if(clear) *net.seen = 0;
if(clear) *net->seen = 0;
char *abase = basecfg(acfgfile);
network anet = parse_network_cfg(acfgfile);
if(aweightfile){
load_weights(&anet, aweightfile);
}
if(clear) *anet.seen = 0;
if(clear) *anet->seen = 0;
int i, j, k;
layer imlayer = {0};
for (i = 0; i < net.n; ++i) {
if (net.layers[i].out_c == 3) {
imlayer = net.layers[i];
for (i = 0; i < net->n; ++i) {
if (net->layers[i].out_c == 3) {
imlayer = net->layers[i];
break;
}
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net.batch*net.subdivisions;
i = *net.seen/imgs;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = net->batch*net->subdivisions;
i = *net->seen/imgs;
data train, buffer;
@ -830,21 +827,21 @@ void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfi
char **paths = (char **)list_to_array(plist);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.d = &buffer;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.size = net.w;
args.min = net->min_crop;
args.max = net->max_crop;
args.angle = net->angle;
args.aspect = net->aspect;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
args.size = net->w;
args.type = CLASSIFICATION_DATA;
args.classes = 1;
char *ls[1] = {"coco"};
@ -856,8 +853,8 @@ void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfi
network_state gstate = {0};
gstate.index = 0;
gstate.net = net;
int x_size = get_network_input_size(net)*net.batch;
int y_size = 1*net.batch;
int x_size = get_network_input_size(net)*net->batch;
int y_size = 1*net->batch;
gstate.input = cuda_make_array(0, x_size);
gstate.truth = 0;
gstate.delta = 0;
@ -868,7 +865,7 @@ void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfi
network_state astate = {0};
astate.index = 0;
astate.net = anet;
int ay_size = get_network_output_size(anet)*anet.batch;
int ay_size = get_network_output_size(anet)*anet->batch;
astate.input = 0;
astate.truth = 0;
astate.delta = 0;
@ -883,7 +880,7 @@ void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfi
//data generated = copy_data(train);
while (get_current_batch(net) < net.max_batches) {
while (get_current_batch(net) < net->max_batches) {
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -896,10 +893,10 @@ void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfi
time=clock();
float gloss = 0;
for(j = 0; j < net.subdivisions; ++j){
get_next_batch(train, net.batch, j*net.batch, X, y);
for(j = 0; j < net->subdivisions; ++j){
get_next_batch(train, net->batch, j*net->batch, X, y);
cuda_push_array(gstate.input, X, x_size);
*net.seen += net.batch;
*net->seen += net->batch;
forward_network_gpu(net, gstate);
fill_gpu(imlayer.outputs, 0, imerror, 1);
@ -917,11 +914,11 @@ void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfi
printf("features %f\n", cuda_mag_array(imlayer.delta_gpu, imlayer.outputs));
printf("realness %f\n", cuda_mag_array(imerror, imlayer.outputs));
gloss += get_network_cost(net) /(net.subdivisions*net.batch);
gloss += get_network_cost(net) /(net->subdivisions*net->batch);
cuda_pull_array(imlayer.output_gpu, imlayer.output, imlayer.outputs*imlayer.batch);
for(k = 0; k < net.batch; ++k){
int index = j*net.batch + k;
for(k = 0; k < net->batch; ++k){
int index = j*net->batch + k;
copy_cpu(imlayer.outputs, imlayer.output + k*imlayer.outputs, 1, generated.X.vals[index], 1);
generated.y.vals[index][0] = 0;
}
@ -977,10 +974,10 @@ void train_lsd(char *cfgfile, char *weightfile, int clear)
if(weightfile){
load_weights(&net, weightfile);
}
if(clear) *net.seen = 0;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net.batch*net.subdivisions;
int i = *net.seen/imgs;
if(clear) *net->seen = 0;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = net->batch*net->subdivisions;
int i = *net->seen/imgs;
data train, buffer;
@ -989,21 +986,21 @@ void train_lsd(char *cfgfile, char *weightfile, int clear)
char **paths = (char **)list_to_array(plist);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.d = &buffer;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.size = net.w;
args.min = net->min_crop;
args.max = net->max_crop;
args.angle = net->angle;
args.aspect = net->aspect;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
args.size = net->w;
args.type = CLASSIFICATION_DATA;
args.classes = 1;
char *ls[1] = {"coco"};
@ -1012,7 +1009,7 @@ void train_lsd(char *cfgfile, char *weightfile, int clear)
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
//while(i*imgs < N*120){
while(get_current_batch(net) < net.max_batches){
while(get_current_batch(net) < net->max_batches){
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -1045,13 +1042,10 @@ void train_lsd(char *cfgfile, char *weightfile, int clear)
}
*/
void test_lsd(char *cfgfile, char *weightfile, char *filename, int gray)
void test_lsd(char *cfg, char *weights, char *filename, int gray)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
srand(2222222);
clock_t time;
@ -1059,8 +1053,8 @@ void test_lsd(char *cfgfile, char *weightfile, char *filename, int gray)
char *input = buff;
int i, imlayer = 0;
for (i = 0; i < net.n; ++i) {
if (net.layers[i].out_c == 3) {
for (i = 0; i < net->n; ++i) {
if (net->layers[i].out_c == 3) {
imlayer = i;
printf("%d\n", i);
break;
@ -1078,8 +1072,8 @@ void test_lsd(char *cfgfile, char *weightfile, char *filename, int gray)
strtok(input, "\n");
}
image im = load_image_color(input, 0, 0);
image resized = resize_min(im, net.w);
image crop = crop_image(resized, (resized.w - net.w)/2, (resized.h - net.h)/2, net.w, net.h);
image resized = resize_min(im, net->w);
image crop = crop_image(resized, (resized.w - net->w)/2, (resized.h - net->h)/2, net->w, net->h);
if(gray) grayscale_image_3c(crop);
float *X = crop.data;

46
examples/nightmare.c
View File

@ -49,14 +49,14 @@ void optimize_picture(network *net, image orig, int max_layer, float scale, floa
net->delta_gpu = cuda_make_array(delta.data, im.w*im.h*im.c);
cuda_push_array(net->input_gpu, im.data, net->inputs);
forward_network_gpu(*net);
forward_network_gpu(net);
copy_gpu(last.outputs, last.output_gpu, 1, last.delta_gpu, 1);
cuda_pull_array(last.delta_gpu, last.delta, last.outputs);
calculate_loss(last.delta, last.delta, last.outputs, thresh);
cuda_push_array(last.delta_gpu, last.delta, last.outputs);
backward_network_gpu(*net);
backward_network_gpu(net);
cuda_pull_array(net->delta_gpu, delta.data, im.w*im.h*im.c);
cuda_free(net->delta_gpu);
@ -64,10 +64,10 @@ void optimize_picture(network *net, image orig, int max_layer, float scale, floa
#else
net->input = im.data;
net->delta = delta.data;
forward_network(*net);
forward_network(net);
copy_cpu(last.outputs, last.output, 1, last.delta, 1);
calculate_loss(last.output, last.delta, last.outputs, thresh);
backward_network(*net);
backward_network(net);
#endif
if(flip) flip_image(delta);
@ -127,7 +127,7 @@ void smooth(image recon, image update, float lambda, int num)
}
}
void reconstruct_picture(network net, float *features, image recon, image update, float rate, float momentum, float lambda, int smooth_size, int iters)
void reconstruct_picture(network *net, float *features, image recon, image update, float rate, float momentum, float lambda, int smooth_size, int iters)
{
int iter = 0;
for (iter = 0; iter < iters; ++iter) {
@ -135,22 +135,22 @@ void reconstruct_picture(network net, float *features, image recon, image update
#ifdef GPU
layer l = get_network_output_layer(net);
cuda_push_array(net.input_gpu, recon.data, recon.w*recon.h*recon.c);
//cuda_push_array(net.truth_gpu, features, net.truths);
net.delta_gpu = cuda_make_array(delta.data, delta.w*delta.h*delta.c);
cuda_push_array(net->input_gpu, recon.data, recon.w*recon.h*recon.c);
//cuda_push_array(net->truth_gpu, features, net->truths);
net->delta_gpu = cuda_make_array(delta.data, delta.w*delta.h*delta.c);
forward_network_gpu(net);
cuda_push_array(l.delta_gpu, features, l.outputs);
axpy_gpu(l.outputs, -1, l.output_gpu, 1, l.delta_gpu, 1);
backward_network_gpu(net);
cuda_pull_array(net.delta_gpu, delta.data, delta.w*delta.h*delta.c);
cuda_pull_array(net->delta_gpu, delta.data, delta.w*delta.h*delta.c);
cuda_free(net.delta_gpu);
cuda_free(net->delta_gpu);
#else
net.input = recon.data;
net.delta = delta.data;
net.truth = features;
net->input = recon.data;
net->delta = delta.data;
net->truth = features;
forward_network(net);
backward_network(net);
@ -206,7 +206,7 @@ void run_lsd(int argc, char **argv)
float *features = 0;
image update;
if (reconstruct){
im = letterbox_image(im, net.w, net.h);
im = letterbox_image(im, net->w, net->h);
int zz = 0;
network_predict(net, im.data);
@ -308,12 +308,12 @@ void run_nightmare(int argc, char **argv)
int reconstruct = find_arg(argc, argv, "-reconstruct");
int smooth_size = find_int_arg(argc, argv, "-smooth", 1);
network net = parse_network_cfg(cfg);
load_weights(&net, weights);
network *net = parse_network_cfg(cfg);
load_weights(net, weights);
char *cfgbase = basecfg(cfg);
char *imbase = basecfg(input);
set_batch_network(&net, 1);
set_batch_network(net, 1);
image im = load_image_color(input, 0, 0);
if(0){
float scale = 1;
@ -325,19 +325,19 @@ void run_nightmare(int argc, char **argv)
free_image(im);
im = resized;
}
//im = letterbox_image(im, net.w, net.h);
//im = letterbox_image(im, net->w, net->h);
float *features = 0;
image update;
if (reconstruct){
net.n = max_layer;
im = letterbox_image(im, net.w, net.h);
net->n = max_layer;
im = letterbox_image(im, net->w, net->h);
//resize_network(&net, im.w, im.h);
network_predict(net, im.data);
if(net.layers[net.n-1].type == REGION){
if(net->layers[net->n-1].type == REGION){
printf("region!\n");
zero_objectness(net.layers[net.n-1]);
zero_objectness(net->layers[net->n-1]);
}
image out_im = copy_image(get_network_image(net));
/*
@ -379,7 +379,7 @@ void run_nightmare(int argc, char **argv)
}else{
int layer = max_layer + rand()%range - range/2;
int octave = rand()%octaves;
optimize_picture(&net, im, layer, 1/pow(1.33333333, octave), rate, thresh, norm);
optimize_picture(net, im, layer, 1/pow(1.33333333, octave), rate, thresh, norm);
}
}
fprintf(stderr, "done\n");

64
examples/regressor.c
View File

@ -10,7 +10,7 @@ void train_regressor(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
char *base = basecfg(cfgfile);
printf("%s\n", base);
printf("%d\n", ngpus);
network *nets = calloc(ngpus, sizeof(network));
network **nets = calloc(ngpus, sizeof(network*));
srand(time(0));
int seed = rand();
@ -19,19 +19,15 @@ void train_regressor(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
#ifdef GPU
cuda_set_device(gpus[i]);
#endif
nets[i] = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&nets[i], weightfile);
}
if(clear) *nets[i].seen = 0;
nets[i].learning_rate *= ngpus;
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i]->learning_rate *= ngpus;
}
srand(time(0));
network net = nets[0];
network *net = nets[0];
int imgs = net.batch * net.subdivisions * ngpus;
int imgs = net->batch * net->subdivisions * ngpus;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
list *options = read_data_cfg(datacfg);
char *backup_directory = option_find_str(options, "backup", "/backup/");
@ -44,18 +40,18 @@ void train_regressor(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
clock_t time;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.threads = 32;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.size = net.w;
args.min = net->min_crop;
args.max = net->max_crop;
args.angle = net->angle;
args.aspect = net->aspect;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
args.size = net->w;
args.paths = paths;
args.n = imgs;
@ -68,8 +64,8 @@ void train_regressor(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
args.d = &buffer;
load_thread = load_data(args);
int epoch = (*net.seen)/N;
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
int epoch = (*net->seen)/N;
while(get_current_batch(net) < net->max_batches || net->max_batches == 0){
time=clock();
pthread_join(load_thread, 0);
@ -91,10 +87,10 @@ void train_regressor(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
#endif
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net->seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net->seen);
free_data(train);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
if(*net->seen/N > epoch){
epoch = *net->seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
save_weights(net, buff);
@ -117,11 +113,8 @@ void train_regressor(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
void predict_regressor(char *cfgfile, char *weightfile, char *filename)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
clock_t time;
@ -138,7 +131,7 @@ void predict_regressor(char *cfgfile, char *weightfile, char *filename)
strtok(input, "\n");
}
image im = load_image_color(input, 0, 0);
image sized = letterbox_image(im, net.w, net.h);
image sized = letterbox_image(im, net->w, net->h);
float *X = sized.data;
time=clock();
@ -156,11 +149,8 @@ void demo_regressor(char *datacfg, char *cfgfile, char *weightfile, int cam_inde
{
#ifdef OPENCV
printf("Regressor Demo\n");
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
CvCapture * cap;
@ -181,7 +171,7 @@ void demo_regressor(char *datacfg, char *cfgfile, char *weightfile, int cam_inde
gettimeofday(&tval_before, NULL);
image in = get_image_from_stream(cap);
image in_s = letterbox_image(in, net.w, net.h);
image in_s = letterbox_image(in, net->w, net->h);
show_image(in, "Regressor");
float *predictions = network_predict(net, in_s.data);

73
examples/rnn.c
View File

@ -171,17 +171,14 @@ void train_char_rnn(char *cfgfile, char *weightfile, char *filename, int clear,
char *base = basecfg(cfgfile);
fprintf(stderr, "%s\n", base);
float avg_loss = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
network *net = load_network(cfgfile, weightfile, clear);
int inputs = net.inputs;
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g, Inputs: %d %d %d\n", net.learning_rate, net.momentum, net.decay, inputs, net.batch, net.time_steps);
int batch = net.batch;
int steps = net.time_steps;
if(clear) *net.seen = 0;
int i = (*net.seen)/net.batch;
int inputs = net->inputs;
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g, Inputs: %d %d %d\n", net->learning_rate, net->momentum, net->decay, inputs, net->batch, net->time_steps);
int batch = net->batch;
int steps = net->time_steps;
if(clear) *net->seen = 0;
int i = (*net->seen)/net->batch;
int streams = batch/steps;
size_t *offsets = calloc(streams, sizeof(size_t));
@ -191,7 +188,7 @@ void train_char_rnn(char *cfgfile, char *weightfile, char *filename, int clear,
}
clock_t time;
while(get_current_batch(net) < net.max_batches){
while(get_current_batch(net) < net->max_batches){
i += 1;
time=clock();
float_pair p;
@ -201,8 +198,8 @@ void train_char_rnn(char *cfgfile, char *weightfile, char *filename, int clear,
p = get_rnn_data(text, offsets, inputs, size, streams, steps);
}
copy_cpu(net.inputs*net.batch, p.x, 1, net.input, 1);
copy_cpu(net.truths*net.batch, p.y, 1, net.truth, 1);
copy_cpu(net->inputs*net->batch, p.x, 1, net->input, 1);
copy_cpu(net->truths*net->batch, p.y, 1, net->truth, 1);
float loss = train_network_datum(net) / (batch);
free(p.x);
free(p.y);
@ -257,14 +254,11 @@ void test_char_rnn(char *cfgfile, char *weightfile, int num, char *seed, float t
char *base = basecfg(cfgfile);
fprintf(stderr, "%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
int inputs = net.inputs;
network *net = load_network(cfgfile, weightfile, 0);
int inputs = net->inputs;
int i, j;
for(i = 0; i < net.n; ++i) net.layers[i].temperature = temp;
for(i = 0; i < net->n; ++i) net->layers[i].temperature = temp;
int c = 0;
int len = strlen(seed);
float *input = calloc(inputs, sizeof(float));
@ -314,14 +308,11 @@ void test_tactic_rnn_multi(char *cfgfile, char *weightfile, int num, float temp,
char *base = basecfg(cfgfile);
fprintf(stderr, "%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
int inputs = net.inputs;
network *net = load_network(cfgfile, weightfile, 0);
int inputs = net->inputs;
int i, j;
for(i = 0; i < net.n; ++i) net.layers[i].temperature = temp;
for(i = 0; i < net->n; ++i) net->layers[i].temperature = temp;
int c = 0;
float *input = calloc(inputs, sizeof(float));
float *out = 0;
@ -362,14 +353,11 @@ void test_tactic_rnn(char *cfgfile, char *weightfile, int num, float temp, int r
char *base = basecfg(cfgfile);
fprintf(stderr, "%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
int inputs = net.inputs;
network *net = load_network(cfgfile, weightfile, 0);
int inputs = net->inputs;
int i, j;
for(i = 0; i < net.n; ++i) net.layers[i].temperature = temp;
for(i = 0; i < net->n; ++i) net->layers[i].temperature = temp;
int c = 0;
float *input = calloc(inputs, sizeof(float));
float *out = 0;
@ -400,11 +388,8 @@ void valid_tactic_rnn(char *cfgfile, char *weightfile, char *seed)
char *base = basecfg(cfgfile);
fprintf(stderr, "%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
int inputs = net.inputs;
network *net = load_network(cfgfile, weightfile, 0);
int inputs = net->inputs;
int count = 0;
int words = 1;
@ -452,11 +437,8 @@ void valid_char_rnn(char *cfgfile, char *weightfile, char *seed)
char *base = basecfg(cfgfile);
fprintf(stderr, "%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
int inputs = net.inputs;
network *net = load_network(cfgfile, weightfile, 0);
int inputs = net->inputs;
int count = 0;
int words = 1;
@ -493,11 +475,8 @@ void vec_char_rnn(char *cfgfile, char *weightfile, char *seed)
char *base = basecfg(cfgfile);
fprintf(stderr, "%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
int inputs = net.inputs;
network *net = load_network(cfgfile, weightfile, 0);
int inputs = net->inputs;
int c;
int seed_len = strlen(seed);
@ -525,7 +504,7 @@ void vec_char_rnn(char *cfgfile, char *weightfile, char *seed)
network_predict(net, input);
input[(int)c] = 0;
layer l = net.layers[0];
layer l = net->layers[0];
#ifdef GPU
cuda_pull_array(l.output_gpu, l.output, l.outputs);
#endif

77
examples/segmenter.c
View File

@ -10,7 +10,7 @@ void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
char *base = basecfg(cfgfile);
printf("%s\n", base);
printf("%d\n", ngpus);
network *nets = calloc(ngpus, sizeof(network));
network **nets = calloc(ngpus, sizeof(network*));
srand(time(0));
int seed = rand();
@ -19,23 +19,20 @@ void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
#ifdef GPU
cuda_set_device(gpus[i]);
#endif
nets[i] = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&nets[i], weightfile);
}
if(clear) *nets[i].seen = 0;
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i]->learning_rate *= ngpus;
}
srand(time(0));
network net = nets[0];
network *net = nets[0];
image pred = get_network_image(net);
int div = net.w/pred.w;
assert(pred.w * div == net.w);
assert(pred.h * div == net.h);
int div = net->w/pred.w;
assert(pred.w * div == net->w);
assert(pred.h * div == net->h);
int imgs = net.batch * net.subdivisions * ngpus;
int imgs = net->batch * net->subdivisions * ngpus;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
list *options = read_data_cfg(datacfg);
char *backup_directory = option_find_str(options, "backup", "/backup/");
@ -48,19 +45,19 @@ void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
clock_t time;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.threads = 32;
args.scale = div;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.size = net.w;
args.min = net->min_crop;
args.max = net->max_crop;
args.angle = net->angle;
args.aspect = net->aspect;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
args.size = net->w;
args.classes = 80;
args.paths = paths;
@ -74,8 +71,8 @@ void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
args.d = &buffer;
load_thread = load_data(args);
int epoch = (*net.seen)/N;
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
int epoch = (*net->seen)/N;
while(get_current_batch(net) < net->max_batches || net->max_batches == 0){
time=clock();
pthread_join(load_thread, 0);
@ -96,8 +93,8 @@ void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
loss = train_network(net, train);
#endif
if(display){
image tr = float_to_image(net.w/div, net.h/div, 80, train.y.vals[net.batch*(net.subdivisions-1)]);
image im = float_to_image(net.w, net.h, net.c, train.X.vals[net.batch*(net.subdivisions-1)]);
image tr = float_to_image(net->w/div, net->h/div, 80, train.y.vals[net->batch*(net->subdivisions-1)]);
image im = float_to_image(net->w, net->h, net->c, train.X.vals[net->batch*(net->subdivisions-1)]);
image mask = mask_to_rgb(tr);
image prmask = mask_to_rgb(pred);
show_image(im, "input");
@ -111,10 +108,10 @@ void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
}
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net->seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net->seen);
free_data(train);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
if(*net->seen/N > epoch){
epoch = *net->seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
save_weights(net, buff);
@ -135,13 +132,10 @@ void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
free(base);
}
void predict_segmenter(char *datafile, char *cfgfile, char *weightfile, char *filename)
void predict_segmenter(char *datafile, char *cfg, char *weights, char *filename)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
srand(2222222);
clock_t time;
@ -158,7 +152,7 @@ void predict_segmenter(char *datafile, char *cfgfile, char *weightfile, char *fi
strtok(input, "\n");
}
image im = load_image_color(input, 0, 0);
image sized = letterbox_image(im, net.w, net.h);
image sized = letterbox_image(im, net->w, net->h);
float *X = sized.data;
time=clock();
@ -180,15 +174,12 @@ void predict_segmenter(char *datafile, char *cfgfile, char *weightfile, char *fi
}
void demo_segmenter(char *datacfg, char *cfgfile, char *weightfile, int cam_index, const char *filename)
void demo_segmenter(char *datacfg, char *cfg, char *weights, int cam_index, const char *filename)
{
#ifdef OPENCV
printf("Classifier Demo\n");
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
srand(2222222);
CvCapture * cap;
@ -209,7 +200,7 @@ void demo_segmenter(char *datacfg, char *cfgfile, char *weightfile, int cam_inde
gettimeofday(&tval_before, NULL);
image in = get_image_from_stream(cap);
image in_s = letterbox_image(in, net.w, net.h);
image in_s = letterbox_image(in, net->w, net->h);
network_predict(net, in_s.data);

27
examples/super.c
View File

@ -8,14 +8,10 @@ void train_super(char *cfgfile, char *weightfile, int clear)
char *base = basecfg(cfgfile);
printf("%s\n", base);
float avg_loss = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
if(clear) *net.seen = 0;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net.batch*net.subdivisions;
int i = *net.seen/imgs;
network *net = load_network(cfgfile, weightfile, clear);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = net->batch*net->subdivisions;
int i = *net->seen/imgs;
data train, buffer;
@ -24,8 +20,8 @@ void train_super(char *cfgfile, char *weightfile, int clear)
char **paths = (char **)list_to_array(plist);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.scale = 4;
args.paths = paths;
args.n = imgs;
@ -36,7 +32,7 @@ void train_super(char *cfgfile, char *weightfile, int clear)
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
//while(i*imgs < N*120){
while(get_current_batch(net) < net.max_batches){
while(get_current_batch(net) < net->max_batches){
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -70,11 +66,8 @@ void train_super(char *cfgfile, char *weightfile, int clear)
void test_super(char *cfgfile, char *weightfile, char *filename)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
clock_t time;
@ -91,7 +84,7 @@ void test_super(char *cfgfile, char *weightfile, char *filename)
strtok(input, "\n");
}
image im = load_image_color(input, 0, 0);
resize_network(&net, im.w, im.h);
resize_network(net, im.w, im.h);
printf("%d %d\n", im.w, im.h);
float *X = im.data;

51
examples/tag.c
View File

@ -7,12 +7,8 @@ void train_tag(char *cfgfile, char *weightfile, int clear)
char *base = basecfg(cfgfile);
char *backup_directory = "/home/pjreddie/backup/";
printf("%s\n", base);
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
if(clear) *net.seen = 0;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfgfile, weightfile, clear);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = 1024;
list *plist = get_paths("/home/pjreddie/tag/train.list");
char **paths = (char **)list_to_array(plist);
@ -24,30 +20,30 @@ void train_tag(char *cfgfile, char *weightfile, int clear)
data buffer;
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.min = net.w;
args.max = net.max_crop;
args.size = net.w;
args.min = net->w;
args.max = net->max_crop;
args.size = net->w;
args.paths = paths;
args.classes = net.outputs;
args.classes = net->outputs;
args.n = imgs;
args.m = N;
args.d = &buffer;
args.type = TAG_DATA;
args.angle = net.angle;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.angle = net->angle;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
fprintf(stderr, "%d classes\n", net.outputs);
fprintf(stderr, "%d classes\n", net->outputs);
load_thread = load_data_in_thread(args);
int epoch = (*net.seen)/N;
while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
int epoch = (*net->seen)/N;
while(get_current_batch(net) < net->max_batches || net->max_batches == 0){
time=clock();
pthread_join(load_thread, 0);
train = buffer;
@ -58,10 +54,10 @@ void train_tag(char *cfgfile, char *weightfile, int clear)
float loss = train_network(net, train);
if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
printf("%ld, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net->seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net->seen);
free_data(train);
if(*net.seen/N > epoch){
epoch = *net.seen/N;
if(*net->seen/N > epoch){
epoch = *net->seen/N;
char buff[256];
sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
save_weights(net, buff);
@ -86,11 +82,8 @@ void train_tag(char *cfgfile, char *weightfile, int clear)
void test_tag(char *cfgfile, char *weightfile, char *filename)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
srand(2222222);
int i = 0;
char **names = get_labels("data/tags.txt");
@ -98,7 +91,7 @@ void test_tag(char *cfgfile, char *weightfile, char *filename)
int indexes[10];
char buff[256];
char *input = buff;
int size = net.w;
int size = net->w;
while(1){
if(filename){
strncpy(input, filename, 256);
@ -111,7 +104,7 @@ void test_tag(char *cfgfile, char *weightfile, char *filename)
}
image im = load_image_color(input, 0, 0);
image r = resize_min(im, size);
resize_network(&net, r.w, r.h);
resize_network(net, r.w, r.h);
printf("%d %d\n", r.w, r.h);
float *X = r.data;

70
examples/yolo.c
View File

@ -10,17 +10,14 @@ void train_yolo(char *cfgfile, char *weightfile)
char *base = basecfg(cfgfile);
printf("%s\n", base);
float avg_loss = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = net.batch*net.subdivisions;
int i = *net.seen/imgs;
network *net = load_network(cfgfile, weightfile, 0);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
int imgs = net->batch*net->subdivisions;
int i = *net->seen/imgs;
data train, buffer;
layer l = net.layers[net.n - 1];
layer l = net->layers[net->n - 1];
int side = l.side;
int classes = l.classes;
@ -31,8 +28,8 @@ void train_yolo(char *cfgfile, char *weightfile)
char **paths = (char **)list_to_array(plist);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.paths = paths;
args.n = imgs;
args.m = plist->size;
@ -42,15 +39,15 @@ void train_yolo(char *cfgfile, char *weightfile)
args.d = &buffer;
args.type = REGION_DATA;
args.angle = net.angle;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
args.angle = net->angle;
args.exposure = net->exposure;
args.saturation = net->saturation;
args.hue = net->hue;
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
//while(i*imgs < N*120){
while(get_current_batch(net) < net.max_batches){
while(get_current_batch(net) < net->max_batches){
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -98,14 +95,11 @@ void print_yolo_detections(FILE **fps, char *id, box *boxes, float **probs, int
}
}
void validate_yolo(char *cfgfile, char *weightfile)
void validate_yolo(char *cfg, char *weights)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
srand(time(0));
char *base = "results/comp4_det_test_";
@ -114,7 +108,7 @@ void validate_yolo(char *cfgfile, char *weightfile)
//list *plist = get_paths("data/voc.2012.test");
char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
int classes = l.classes;
int j;
@ -144,8 +138,8 @@ void validate_yolo(char *cfgfile, char *weightfile)
pthread_t *thr = calloc(nthreads, sizeof(pthread_t));
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.w = net->w;
args.h = net->h;
args.type = IMAGE_DATA;
for(t = 0; t < nthreads; ++t){
@ -186,21 +180,18 @@ void validate_yolo(char *cfgfile, char *weightfile)
fprintf(stderr, "Total Detection Time: %f Seconds\n", (double)(time(0) - start));
}
void validate_yolo_recall(char *cfgfile, char *weightfile)
void validate_yolo_recall(char *cfg, char *weights)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
network *net = load_network(cfg, weights, 0);
set_batch_network(net, 1);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
srand(time(0));
char *base = "results/comp4_det_test_";
list *plist = get_paths("data/voc.2007.test");
char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
int classes = l.classes;
int side = l.side;
@ -230,7 +221,7 @@ void validate_yolo_recall(char *cfgfile, char *weightfile)
for(i = 0; i < m; ++i){
char *path = paths[i];
image orig = load_image_color(path, 0, 0);
image sized = resize_image(orig, net.w, net.h);
image sized = resize_image(orig, net->w, net->h);
char *id = basecfg(path);
network_predict(net, sized.data);
get_detection_boxes(l, orig.w, orig.h, thresh, probs, boxes, 1);
@ -275,12 +266,9 @@ void validate_yolo_recall(char *cfgfile, char *weightfile)
void test_yolo(char *cfgfile, char *weightfile, char *filename, float thresh)
{
image **alphabet = load_alphabet();
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
layer l = net.layers[net.n-1];
set_batch_network(&net, 1);
network *net = load_network(cfgfile, weightfile, 0);
layer l = net->layers[net->n-1];
set_batch_network(net, 1);
srand(2222222);
clock_t time;
char buff[256];
@ -301,7 +289,7 @@ void test_yolo(char *cfgfile, char *weightfile, char *filename, float thresh)
strtok(input, "\n");
}
image im = load_image_color(input,0,0);
image sized = resize_image(im, net.w, net.h);
image sized = resize_image(im, net->w, net->h);
float *X = sized.data;
time=clock();
network_predict(net, X);

70
include/darknet.h
View File

@ -446,12 +446,15 @@ typedef struct network{
int h, w, c;
int max_crop;
int min_crop;
float max_ratio;
float min_ratio;
int center;
float angle;
float aspect;
float exposure;
float saturation;
float hue;
int random;
int gpu_index;
tree *hierarchy;
@ -553,9 +556,8 @@ typedef struct{
} box_label;
network load_network(char *cfg, char *weights, int clear);
network *load_network_p(char *cfg, char *weights, int clear);
load_args get_base_args(network net);
network *load_network(char *cfg, char *weights, int clear);
load_args get_base_args(network *net);
void free_data(data d);
@ -575,10 +577,11 @@ pthread_t load_data(load_args args);
list *read_data_cfg(char *filename);
list *read_cfg(char *filename);
unsigned char *read_file(char *filename);
data resize_data(data orig, int w, int h);
void forward_network(network net);
void backward_network(network net);
void update_network(network net);
void forward_network(network *net);
void backward_network(network *net);
void update_network(network *net);
void axpy_cpu(int N, float ALPHA, float *X, int INCX, float *Y, int INCY);
@ -600,20 +603,20 @@ void cuda_pull_array(float *x_gpu, float *x, size_t n);
float cuda_mag_array(float *x_gpu, size_t n);
void cuda_push_array(float *x_gpu, float *x, size_t n);
void forward_network_gpu(network net);
void backward_network_gpu(network net);
void update_network_gpu(network net);
void forward_network_gpu(network *net);
void backward_network_gpu(network *net);
void update_network_gpu(network *net);
float train_networks(network *nets, int n, data d, int interval);
void sync_nets(network *nets, int n, int interval);
void harmless_update_network_gpu(network net);
float train_networks(network **nets, int n, data d, int interval);
void sync_nets(network **nets, int n, int interval);
void harmless_update_network_gpu(network *net);
#endif
void save_image_png(image im, const char *name);
void get_next_batch(data d, int n, int offset, float *X, float *y);
void grayscale_image_3c(image im);
void normalize_image(image p);
void matrix_to_csv(matrix m);
float train_network_sgd(network net, data d, int n);
float train_network_sgd(network *net, data d, int n);
void rgbgr_image(image im);
data copy_data(data d);
data concat_data(data d1, data d2);
@ -622,8 +625,8 @@ float matrix_topk_accuracy(matrix truth, matrix guess, int k);
void matrix_add_matrix(matrix from, matrix to);
void scale_matrix(matrix m, float scale);
matrix csv_to_matrix(char *filename);
float *network_accuracies(network net, data d, int n);
float train_network_datum(network net);
float *network_accuracies(network *net, data d, int n);
float train_network_datum(network *net);
image make_random_image(int w, int h, int c);
void denormalize_connected_layer(layer l);
@ -639,17 +642,17 @@ void get_detection_boxes(layer l, int w, int h, float thresh, float **probs, box
char *option_find_str(list *l, char *key, char *def);
int option_find_int(list *l, char *key, int def);
network parse_network_cfg(char *filename);
void save_weights(network net, char *filename);
network *parse_network_cfg(char *filename);
void save_weights(network *net, char *filename);
void load_weights(network *net, char *filename);
void save_weights_upto(network net, char *filename, int cutoff);
void save_weights_upto(network *net, char *filename, int cutoff);
void load_weights_upto(network *net, char *filename, int start, int cutoff);
void zero_objectness(layer l);
void get_region_boxes(layer l, int w, int h, int netw, int neth, float thresh, float **probs, box *boxes, float **masks, int only_objectness, int *map, float tree_thresh, int relative);
void free_network(network net);
void free_network(network *net);
void set_batch_network(network *net, int b);
void set_temp_network(network net, float t);
void set_temp_network(network *net, float t);
image load_image(char *filename, int w, int h, int c);
image load_image_color(char *filename, int w, int h);
image make_image(int w, int h, int c);
@ -657,6 +660,7 @@ image resize_image(image im, int w, int h);
image letterbox_image(image im, int w, int h);
image crop_image(image im, int dx, int dy, int w, int h);
image resize_min(image im, int min);
image resize_max(image im, int max);
image threshold_image(image im, float thresh);
image mask_to_rgb(image mask);
int resize_network(network *net, int w, int h);
@ -666,25 +670,25 @@ void save_image(image p, const char *name);
void show_image(image p, const char *name);
image copy_image(image p);
void draw_box_width(image a, int x1, int y1, int x2, int y2, int w, float r, float g, float b);
float get_current_rate(network net);
float get_current_rate(network *net);
void composite_3d(char *f1, char *f2, char *out, int delta);
data load_data_old(char **paths, int n, int m, char **labels, int k, int w, int h);
size_t get_current_batch(network net);
size_t get_current_batch(network *net);
void constrain_image(image im);
image get_network_image_layer(network net, int i);
layer get_network_output_layer(network net);
void top_predictions(network net, int n, int *index);
image get_network_image_layer(network *net, int i);
layer get_network_output_layer(network *net);
void top_predictions(network *net, int n, int *index);
void flip_image(image a);
image float_to_image(int w, int h, int c, float *data);
void ghost_image(image source, image dest, int dx, int dy);
float network_accuracy(network net, data d);
float network_accuracy(network *net, data d);
void random_distort_image(image im, float hue, float saturation, float exposure);
void fill_image(image m, float s);
image grayscale_image(image im);
void rotate_image_cw(image im, int times);
double what_time_is_it_now();
image rotate_image(image m, float rad);
void visualize_network(network net);
void visualize_network(network *net);
float box_iou(box a, box b);
void do_nms(box *boxes, float **probs, int total, int classes, float thresh);
data load_all_cifar10();
@ -692,11 +696,10 @@ box_label *read_boxes(char *filename, int *n);
box float_to_box(float *f, int stride);
void draw_detections(image im, int num, float thresh, box *boxes, float **probs, float **masks, char **names, image **alphabet, int classes);
matrix network_predict_data(network net, data test);
matrix network_predict_data(network *net, data test);
image **load_alphabet();
image get_network_image(network net);
float *network_predict(network net, float *input);
float *network_predict_p(network *net, float *input);
image get_network_image(network *net);
float *network_predict(network *net, float *input);
int network_width(network *net);
int network_height(network *net);
@ -705,8 +708,7 @@ void network_detect(network *net, image im, float thresh, float hier_thresh, flo
int num_boxes(network *net);
box *make_boxes(network *net);
void reset_network_state(network net, int b);
void reset_network_state(network net, int b);
void reset_network_state(network *net, int b);
char **get_labels(char *filename);
void do_nms_sort(box *boxes, float **probs, int total, int classes, float thresh);
@ -720,7 +722,7 @@ image get_image_from_stream(CvCapture *cap);
#endif
#endif
void free_image(image m);
float train_network(network net, data d);
float train_network(network *net, data d);
pthread_t load_data_in_thread(load_args args);
void load_data_blocking(load_args args);
list *get_paths(char *filename);

6
python/darknet.py
View File

@ -38,7 +38,7 @@ lib.network_width.restype = c_int
lib.network_height.argtypes = [c_void_p]
lib.network_height.restype = c_int
predict = lib.network_predict_p
predict = lib.network_predict
predict.argtypes = [c_void_p, POINTER(c_float)]
predict.restype = POINTER(c_float)
@ -57,13 +57,13 @@ make_probs = lib.make_probs
make_probs.argtypes = [c_void_p]
make_probs.restype = POINTER(POINTER(c_float))
detect = lib.network_predict_p
detect = lib.network_predict
detect.argtypes = [c_void_p, IMAGE, c_float, c_float, c_float, POINTER(BOX), POINTER(POINTER(c_float))]
reset_rnn = lib.reset_rnn
reset_rnn.argtypes = [c_void_p]
load_net = lib.load_network_p
load_net = lib.load_network
load_net.argtypes = [c_char_p, c_char_p, c_int]
load_net.restype = c_void_p

21
src/data.c
View File

@ -1172,11 +1172,32 @@ data load_data_regression(char **paths, int n, int m, int min, int max, int size
return d;
}
data resize_data(data orig, int w, int h)
{
data d = {0};
d.shallow = 0;
d.w = w;
d.h = h;
int i;
d.X.rows = orig.X.rows;
d.X.cols = w*h*3;
d.X.vals = calloc(d.X.rows, sizeof(float));
d.y = copy_matrix(orig.y);
for(i = 0; i < orig.X.rows; ++i){
image im = float_to_image(orig.w, orig.h, 3, orig.X.vals[i]);
d.X.vals[i] = resize_image(im, w, h).data;
}
return d;
}
data load_data_augment(char **paths, int n, int m, char **labels, int k, tree *hierarchy, int min, int max, int size, float angle, float aspect, float hue, float saturation, float exposure, int center)
{
if(m) paths = get_random_paths(paths, n, m);
data d = {0};
d.shallow = 0;
d.w=size;
d.h=size;
d.X = load_image_augment_paths(paths, n, min, max, size, angle, aspect, hue, saturation, exposure, center);
d.y = load_labels_paths(paths, n, labels, k, hierarchy);
if(m) free(paths);

33
src/demo.c
View File

@ -19,7 +19,7 @@ static int demo_classes;
static float **probs;
static box *boxes;
static network net;
static network *net;
static image buff [3];
static image buff_letter[3];
static int buff_index = 0;
@ -43,7 +43,7 @@ void *detect_in_thread(void *ptr)
running = 1;
float nms = .4;
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
float *X = buff_letter[(buff_index+2)%3].data;
float *prediction = network_predict(net, X);
@ -53,7 +53,7 @@ void *detect_in_thread(void *ptr)
if(l.type == DETECTION){
get_detection_boxes(l, 1, 1, demo_thresh, probs, boxes, 0);
} else if (l.type == REGION){
get_region_boxes(l, buff[0].w, buff[0].h, net.w, net.h, demo_thresh, probs, boxes, 0, 0, 0, demo_hier, 1);
get_region_boxes(l, buff[0].w, buff[0].h, net->w, net->h, demo_thresh, probs, boxes, 0, 0, 0, demo_hier, 1);
} else {
error("Last layer must produce detections\n");
}
@ -74,7 +74,7 @@ void *detect_in_thread(void *ptr)
void *fetch_in_thread(void *ptr)
{
int status = fill_image_from_stream(cap, buff[buff_index]);
letterbox_image_into(buff[buff_index], net.w, net.h, buff_letter[buff_index]);
letterbox_image_into(buff[buff_index], net->w, net->h, buff_letter[buff_index]);
if(status == 0) demo_done = 1;
return 0;
}
@ -126,11 +126,8 @@ void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const ch
demo_thresh = thresh;
demo_hier = hier;
printf("Demo\n");
net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
net = load_network(cfgfile, weightfile, 0);
set_batch_network(net, 1);
pthread_t detect_thread;
pthread_t fetch_thread;
@ -155,7 +152,7 @@ void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const ch
if(!cap) error("Couldn't connect to webcam.\n");
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
demo_detections = l.n*l.w*l.h;
int j;
@ -169,9 +166,9 @@ void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const ch
buff[0] = get_image_from_stream(cap);
buff[1] = copy_image(buff[0]);
buff[2] = copy_image(buff[0]);
buff_letter[0] = letterbox_image(buff[0], net.w, net.h);
buff_letter[1] = letterbox_image(buff[0], net.w, net.h);
buff_letter[2] = letterbox_image(buff[0], net.w, net.h);
buff_letter[0] = letterbox_image(buff[0], net->w, net->h);
buff_letter[1] = letterbox_image(buff[0], net->w, net->h);
buff_letter[2] = letterbox_image(buff[0], net->w, net->h);
ipl = cvCreateImage(cvSize(buff[0].w,buff[0].h), IPL_DEPTH_8U, buff[0].c);
int count = 0;
@ -218,7 +215,7 @@ void demo_compare(char *cfg1, char *weight1, char *cfg2, char *weight2, float th
demo_hier = hier;
printf("Demo\n");
net = load_network(cfg1, weight1, 0);
set_batch_network(&net, 1);
set_batch_network(net, 1);
pthread_t detect_thread;
pthread_t fetch_thread;
@ -243,7 +240,7 @@ void demo_compare(char *cfg1, char *weight1, char *cfg2, char *weight2, float th
if(!cap) error("Couldn't connect to webcam.\n");
layer l = net.layers[net.n-1];
layer l = net->layers[net->n-1];
demo_detections = l.n*l.w*l.h;
int j;
@ -257,9 +254,9 @@ void demo_compare(char *cfg1, char *weight1, char *cfg2, char *weight2, float th
buff[0] = get_image_from_stream(cap);
buff[1] = copy_image(buff[0]);
buff[2] = copy_image(buff[0]);
buff_letter[0] = letterbox_image(buff[0], net.w, net.h);
buff_letter[1] = letterbox_image(buff[0], net.w, net.h);
buff_letter[2] = letterbox_image(buff[0], net.w, net.h);
buff_letter[0] = letterbox_image(buff[0], net->w, net->h);
buff_letter[1] = letterbox_image(buff[0], net->w, net->h);
buff_letter[2] = letterbox_image(buff[0], net->w, net->h);
ipl = cvCreateImage(cvSize(buff[0].w,buff[0].h), IPL_DEPTH_8U, buff[0].c);
int count = 0;

681
src/network.c
View File

@ -30,64 +30,46 @@
#include "parser.h"
#include "data.h"
load_args get_base_args(network net)
load_args get_base_args(network *net)
{
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.size = net.w;
args.w = net->w;
args.h = net->h;
args.size = net->w;
args.min = net.min_crop;
args.max = net.max_crop;
args.angle = net.angle;
args.aspect = net.aspect;
args.exposure = net.exposure;
args.center = net.center;
args.saturation = net.saturation;
args.hue = net.hue;
args.min = net->min_crop;
args.max = net->max_crop;
args.angle = net->angle;
args.aspect = net->aspect;
args.exposure = net->exposure;
args.center = net->center;
args.saturation = net->saturation;
args.hue = net->hue;
return args;
}
network load_network(char *cfg, char *weights, int clear)
network *load_network(char *cfg, char *weights, int clear)
{
network net = parse_network_cfg(cfg);
network *net = parse_network_cfg(cfg);
if(weights && weights[0] != 0){
load_weights(&net, weights);
load_weights(net, weights);
}
if(clear) *net.seen = 0;
if(clear) (*net->seen) = 0;
return net;
}
network *load_network_p(char *cfg, char *weights, int clear)
size_t get_current_batch(network *net)
{
network *net = calloc(1, sizeof(network));
*net = load_network(cfg, weights, clear);
return net;
}
size_t get_current_batch(network net)
{
size_t batch_num = (*net.seen)/(net.batch*net.subdivisions);
size_t 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(net.gpu_index >= 0) update_network_gpu(net);
#endif
}
void reset_network_state(network net, int b)
void reset_network_state(network *net, int b)
{
int i;
for (i = 0; i < net.n; ++i) {
for (i = 0; i < net->n; ++i) {
#ifdef GPU
layer l = net.layers[i];
layer l = net->layers[i];
if(l.state_gpu){
fill_gpu(l.outputs, 0, l.state_gpu + l.outputs*b, 1);
}
@ -100,39 +82,38 @@ void reset_network_state(network net, int b)
void reset_rnn(network *net)
{
reset_network_state(*net, 0);
reset_network_state(net, 0);
}
float get_current_rate(network net)
float get_current_rate(network *net)
{
size_t batch_num = get_current_batch(net);
int i;
float rate;
if (batch_num < net.burn_in) return net.learning_rate * pow((float)batch_num / net.burn_in, net.power);
switch (net.policy) {
if (batch_num < net->burn_in) return net->learning_rate * pow((float)batch_num / net->burn_in, net->power);
switch (net->policy) {
case CONSTANT:
return net.learning_rate;
return net->learning_rate;
case STEP:
return net.learning_rate * pow(net.scale, batch_num/net.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 && net.scales[i] > 1) reset_momentum(net);
rate = net->learning_rate;
for(i = 0; i < net->num_steps; ++i){
if(net->steps[i] > batch_num) return rate;
rate *= net->scales[i];
}
return rate;
case EXP:
return net.learning_rate * pow(net.gamma, batch_num);
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);
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);
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))));
return net->learning_rate * (1./(1.+exp(net->gamma*(batch_num - net->step))));
default:
fprintf(stderr, "Policy is weird!\n");
return net.learning_rate;
return net->learning_rate;
}
}
@ -189,19 +170,26 @@ char *get_layer_string(LAYER_TYPE a)
return "none";
}
network make_network(int n)
network *make_network(int n)
{
network net = {0};
net.n = n;
net.layers = calloc(net.n, sizeof(layer));
net.seen = calloc(1, sizeof(size_t));
net.t = calloc(1, sizeof(int));
net.cost = calloc(1, sizeof(float));
network *net = calloc(1, sizeof(network));
net->n = n;
net->layers = calloc(net->n, sizeof(layer));
net->seen = calloc(1, sizeof(size_t));
net->t = calloc(1, sizeof(int));
net->cost = calloc(1, sizeof(float));
return net;
}
void forward_network(network net)
void forward_network(network *netp)
{
#ifdef GPU
if(netp->gpu_index >= 0){
forward_network_gpu(netp);
return;
}
#endif
network net = *netp;
int i;
for(i = 0; i < net.n; ++i){
net.index = i;
@ -215,15 +203,22 @@ void forward_network(network net)
net.truth = l.output;
}
}
calc_network_cost(net);
calc_network_cost(netp);
}
void update_network(network net)
void update_network(network *netp)
{
#ifdef GPU
if(netp->gpu_index >= 0){
update_network_gpu(netp);
return;
}
#endif
network net = *netp;
int i;
update_args a = {0};
a.batch = net.batch*net.subdivisions;
a.learning_rate = get_current_rate(net);
a.learning_rate = get_current_rate(netp);
a.momentum = net.momentum;
a.decay = net.decay;
a.adam = net.adam;
@ -241,8 +236,9 @@ void update_network(network net)
}
}
void calc_network_cost(network net)
void calc_network_cost(network *netp)
{
network net = *netp;
int i;
float sum = 0;
int count = 0;
@ -255,13 +251,20 @@ void calc_network_cost(network net)
*net.cost = sum/count;
}
int get_predicted_class_network(network net)
int get_predicted_class_network(network *net)
{
return max_index(net.output, net.outputs);
return max_index(net->output, net->outputs);
}
void backward_network(network net)
void backward_network(network *netp)
{
#ifdef GPU
if(netp->gpu_index >= 0){
backward_network_gpu(netp);
return;
}
#endif
network net = *netp;
int i;
network orig = net;
for(i = net.n-1; i >= 0; --i){
@ -279,55 +282,52 @@ void backward_network(network net)
}
}
float train_network_datum(network net)
float train_network_datum(network *net)
{
#ifdef GPU
if(gpu_index >= 0) return train_network_datum_gpu(net);
#endif
*net.seen += net.batch;
net.train = 1;
*net->seen += net->batch;
net->train = 1;
forward_network(net);
backward_network(net);
float error = *net.cost;
if(((*net.seen)/net.batch)%net.subdivisions == 0) update_network(net);
float error = *net->cost;
if(((*net->seen)/net->batch)%net->subdivisions == 0) update_network(net);
return error;
}
float train_network_sgd(network net, data d, int n)
float train_network_sgd(network *net, data d, int n)
{
int batch = net.batch;
int batch = net->batch;
int i;
float sum = 0;
for(i = 0; i < n; ++i){
get_random_batch(d, batch, net.input, net.truth);
get_random_batch(d, batch, net->input, net->truth);
float err = train_network_datum(net);
sum += err;
}
return (float)sum/(n*batch);
}
float train_network(network net, data d)
float train_network(network *net, data d)
{
assert(d.X.rows % net.batch == 0);
int batch = net.batch;
assert(d.X.rows % net->batch == 0);
int batch = net->batch;
int n = d.X.rows / batch;
int i;
float sum = 0;
for(i = 0; i < n; ++i){
get_next_batch(d, batch, i*batch, net.input, net.truth);
get_next_batch(d, batch, i*batch, net->input, net->truth);
float err = train_network_datum(net);
sum += err;
}
return (float)sum/(n*batch);
}
void set_temp_network(network net, float t)
void set_temp_network(network *net, float t)
{
int i;
for(i = 0; i < net.n; ++i){
net.layers[i].temperature = t;
for(i = 0; i < net->n; ++i){
net->layers[i].temperature = t;
}
}
@ -395,7 +395,7 @@ int resize_network(network *net, int w, int h)
h = l.out_h;
if(l.type == AVGPOOL) break;
}
layer out = get_network_output_layer(*net);
layer out = get_network_output_layer(net);
net->inputs = net->layers[0].inputs;
net->outputs = out.outputs;
net->truths = out.outputs;
@ -424,22 +424,22 @@ int resize_network(network *net, int w, int h)
return 0;
}
detection_layer get_network_detection_layer(network net)
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];
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};
layer l = {0};
return l;
}
image get_network_image_layer(network net, int i)
image get_network_image_layer(network *net, int i)
{
layer l = net.layers[i];
layer l = net->layers[i];
#ifdef GPU
//cuda_pull_array(l.output_gpu, l.output, l.outputs);
#endif
@ -450,10 +450,10 @@ image get_network_image_layer(network net, int i)
return def;
}
image get_network_image(network net)
image get_network_image(network *net)
{
int i;
for(i = net.n-1; i >= 0; --i){
for(i = net->n-1; i >= 0; --i){
image m = get_network_image_layer(net, i);
if(m.h != 0) return m;
}
@ -461,37 +461,37 @@ image get_network_image(network net)
return def;
}
void visualize_network(network net)
void visualize_network(network *net)
{
image *prev = 0;
int i;
char buff[256];
for(i = 0; i < net.n; ++i){
for(i = 0; i < net->n; ++i){
sprintf(buff, "Layer %d", i);
layer l = net.layers[i];
layer l = net->layers[i];
if(l.type == CONVOLUTIONAL){
prev = visualize_convolutional_layer(l, buff, prev);
}
}
}
void top_predictions(network net, int k, int *index)
void top_predictions(network *net, int k, int *index)
{
top_k(net.output, net.outputs, k, index);
top_k(net->output, net->outputs, k, index);
}
float *network_predict(network net, float *input)
float *network_predict(network *net, float *input)
{
#ifdef GPU
if(gpu_index >= 0) return network_predict_gpu(net, input);
#endif
net.input = input;
net.truth = 0;
net.train = 0;
net.delta = 0;
network orig = *net;
net->input = input;
net->truth = 0;
net->train = 0;
net->delta = 0;
forward_network(net);
return net.output;
float *out = net->output;
*net = orig;
return out;
}
int num_boxes(network *net)
@ -526,16 +526,11 @@ void network_detect(network *net, image im, float thresh, float hier_thresh, flo
}
}
float *network_predict_p(network *net, float *input)
{
return network_predict(*net, input);
}
float *network_predict_image(network *net, image im)
{
image imr = letterbox_image(im, net->w, net->h);
set_batch_network(net, 1);
float *p = network_predict(*net, imr.data);
float *p = network_predict(net, imr.data);
free_image(imr);
return p;
}
@ -543,20 +538,20 @@ float *network_predict_image(network *net, image im)
int network_width(network *net){return net->w;}
int network_height(network *net){return net->h;}
matrix network_predict_data_multi(network net, data test, int n)
matrix network_predict_data_multi(network *net, data test, int n)
{
int i,j,b,m;
int k = net.outputs;
int k = net->outputs;
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){
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){
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;
@ -568,19 +563,19 @@ matrix network_predict_data_multi(network net, data test, int n)
return pred;
}
matrix network_predict_data(network net, data test)
matrix network_predict_data(network *net, data test)
{
int i,j,b;
int k = net.outputs;
int k = net->outputs;
matrix pred = make_matrix(test.X.rows, k);
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){
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;
memcpy(X+b*test.X.cols, test.X.vals[i+b], test.X.cols*sizeof(float));
}
float *out = network_predict(net, X);
for(b = 0; b < net.batch; ++b){
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];
@ -591,11 +586,11 @@ matrix network_predict_data(network net, data test)
return pred;
}
void print_network(network net)
void print_network(network *net)
{
int i,j;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
for(i = 0; i < net->n; ++i){
layer l = net->layers[i];
float *output = l.output;
int n = l.outputs;
float mean = mean_array(output, n);
@ -608,7 +603,7 @@ void print_network(network net)
}
}
void compare_networks(network n1, network n2, data test)
void compare_networks(network *n1, network *n2, data test)
{
matrix g1 = network_predict_data(n1, test);
matrix g2 = network_predict_data(n2, test);
@ -633,7 +628,7 @@ void compare_networks(network n1, network n2, data test)
printf("%f\n", num/den);
}
float network_accuracy(network net, data d)
float network_accuracy(network *net, data d)
{
matrix guess = network_predict_data(net, d);
float acc = matrix_topk_accuracy(d.y, guess,1);
@ -641,7 +636,7 @@ float network_accuracy(network net, data d)
return acc;
}
float *network_accuracies(network net, data d, int n)
float *network_accuracies(network *net, data d, int n)
{
static float acc[2];
matrix guess = network_predict_data(net, d);
@ -651,16 +646,16 @@ float *network_accuracies(network net, data d, int n)
return acc;
}
layer get_network_output_layer(network net)
layer get_network_output_layer(network *net)
{
int i;
for(i = net.n - 1; i >= 0; --i){
if(net.layers[i].type != COST) break;
for(i = net->n - 1; i >= 0; --i){
if(net->layers[i].type != COST) break;
}
return net.layers[i];
return net->layers[i];
}
float network_accuracy_multi(network net, data d, int n)
float network_accuracy_multi(network *net, data d, int n)
{
matrix guess = network_predict_data_multi(net, d, n);
float acc = matrix_topk_accuracy(d.y, guess,1);
@ -668,45 +663,417 @@ float network_accuracy_multi(network net, data d, int n)
return acc;
}
void free_network(network net)
void free_network(network *net)
{
int i;
for(i = 0; i < net.n; ++i){
free_layer(net.layers[i]);
for(i = 0; i < net->n; ++i){
free_layer(net->layers[i]);
}
free(net.layers);
if(net.input) free(net.input);
if(net.truth) free(net.truth);
free(net->layers);
if(net->input) free(net->input);
if(net->truth) free(net->truth);
#ifdef GPU
if(net.input_gpu) cuda_free(net.input_gpu);
if(net.truth_gpu) cuda_free(net.truth_gpu);
if(net->input_gpu) cuda_free(net->input_gpu);
if(net->truth_gpu) cuda_free(net->truth_gpu);
#endif
free(net);
}
// Some day...
// ^ What the hell is this comment for?
layer network_output_layer(network net)
layer network_output_layer(network *net)
{
int i;
for(i = net.n - 1; i >= 0; --i){
if(net.layers[i].type != COST) break;
for(i = net->n - 1; i >= 0; --i){
if(net->layers[i].type != COST) break;
}
return net.layers[i];
return net->layers[i];
}
int network_inputs(network net)
int network_inputs(network *net)
{
return net.layers[0].inputs;
return net->layers[0].inputs;
}
int network_outputs(network net)
int network_outputs(network *net)
{
return network_output_layer(net).outputs;
}
float *network_output(network net)
float *network_output(network *net)
{
return network_output_layer(net).output;
}
#ifdef GPU
void forward_network_gpu(network *netp)
{
network net = *netp;
cuda_set_device(net.gpu_index);
cuda_push_array(net.input_gpu, net.input, net.inputs*net.batch);
if(net.truth){
cuda_push_array(net.truth_gpu, net.truth, net.truths*net.batch);
}
int i;
for(i = 0; i < net.n; ++i){
net.index = i;
layer l = net.layers[i];
if(l.delta_gpu){
fill_gpu(l.outputs * l.batch, 0, l.delta_gpu, 1);
}
l.forward_gpu(l, net);
net.input_gpu = l.output_gpu;
net.input = l.output;
if(l.truth) {
net.truth_gpu = l.output_gpu;
net.truth = l.output;
}
}
pull_network_output(netp);
calc_network_cost(netp);
}
void backward_network_gpu(network *netp)
{
int i;
network net = *netp;
network orig = net;
cuda_set_device(net.gpu_index);
for(i = net.n-1; i >= 0; --i){
layer l = net.layers[i];
if(l.stopbackward) break;
if(i == 0){
net = orig;
}else{
layer prev = net.layers[i-1];
net.input = prev.output;
net.delta = prev.delta;
net.input_gpu = prev.output_gpu;
net.delta_gpu = prev.delta_gpu;
}
net.index = i;
l.backward_gpu(l, net);
}
}
void update_network_gpu(network *netp)
{
network net = *netp;
cuda_set_device(net.gpu_index);
int i;
update_args a = {0};
a.batch = net.batch*net.subdivisions;
a.learning_rate = get_current_rate(netp);
a.momentum = net.momentum;
a.decay = net.decay;
a.adam = net.adam;
a.B1 = net.B1;
a.B2 = net.B2;
a.eps = net.eps;
++*net.t;
a.t = (*net.t);
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
if(l.update_gpu){
l.update_gpu(l, a);
}
}
}
void harmless_update_network_gpu(network *netp)
{
network net = *netp;
cuda_set_device(net.gpu_index);
int i;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
if(l.weight_updates_gpu) fill_gpu(l.nweights, 0, l.weight_updates_gpu, 1);
if(l.bias_updates_gpu) fill_gpu(l.nbiases, 0, l.bias_updates_gpu, 1);
if(l.scale_updates_gpu) fill_gpu(l.nbiases, 0, l.scale_updates_gpu, 1);
}
}
typedef struct {
network *net;
data d;
float *err;
} train_args;
void *train_thread(void *ptr)
{
train_args args = *(train_args*)ptr;
free(ptr);
cuda_set_device(args.net->gpu_index);
*args.err = train_network(args.net, args.d);
return 0;
}
pthread_t train_network_in_thread(network *net, data d, float *err)
{
pthread_t thread;
train_args *ptr = (train_args *)calloc(1, sizeof(train_args));
ptr->net = net;
ptr->d = d;
ptr->err = err;
if(pthread_create(&thread, 0, train_thread, ptr)) error("Thread creation failed");
return thread;
}
void merge_weights(layer l, layer base)
{
if (l.type == CONVOLUTIONAL) {
axpy_cpu(l.n, 1, l.bias_updates, 1, base.biases, 1);
axpy_cpu(l.nweights, 1, l.weight_updates, 1, base.weights, 1);
if (l.scales) {
axpy_cpu(l.n, 1, l.scale_updates, 1, base.scales, 1);
}
} else if(l.type == CONNECTED) {
axpy_cpu(l.outputs, 1, l.bias_updates, 1, base.biases, 1);
axpy_cpu(l.outputs*l.inputs, 1, l.weight_updates, 1, base.weights, 1);
}
}
void scale_weights(layer l, float s)
{
if (l.type == CONVOLUTIONAL) {
scal_cpu(l.n, s, l.biases, 1);
scal_cpu(l.nweights, s, l.weights, 1);
if (l.scales) {
scal_cpu(l.n, s, l.scales, 1);
}
} else if(l.type == CONNECTED) {
scal_cpu(l.outputs, s, l.biases, 1);
scal_cpu(l.outputs*l.inputs, s, l.weights, 1);
}
}
void pull_weights(layer l)
{
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
cuda_pull_array(l.biases_gpu, l.bias_updates, l.n);
cuda_pull_array(l.weights_gpu, l.weight_updates, l.nweights);
if(l.scales) cuda_pull_array(l.scales_gpu, l.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_pull_array(l.biases_gpu, l.bias_updates, l.outputs);
cuda_pull_array(l.weights_gpu, l.weight_updates, l.outputs*l.inputs);
}
}
void push_weights(layer l)
{
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
cuda_push_array(l.biases_gpu, l.biases, l.n);
cuda_push_array(l.weights_gpu, l.weights, l.nweights);
if(l.scales) cuda_push_array(l.scales_gpu, l.scales, l.n);
} else if(l.type == CONNECTED){
cuda_push_array(l.biases_gpu, l.biases, l.outputs);
cuda_push_array(l.weights_gpu, l.weights, l.outputs*l.inputs);
}
}
void distribute_weights(layer l, layer base)
{
if (l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL) {
cuda_push_array(l.biases_gpu, base.biases, l.n);
cuda_push_array(l.weights_gpu, base.weights, l.nweights);
if (base.scales) cuda_push_array(l.scales_gpu, base.scales, l.n);
} else if (l.type == CONNECTED) {
cuda_push_array(l.biases_gpu, base.biases, l.outputs);
cuda_push_array(l.weights_gpu, base.weights, l.outputs*l.inputs);
}
}
/*
void pull_updates(layer l)
{
if(l.type == CONVOLUTIONAL){
cuda_pull_array(l.bias_updates_gpu, l.bias_updates, l.n);
cuda_pull_array(l.weight_updates_gpu, l.weight_updates, l.nweights);
if(l.scale_updates) cuda_pull_array(l.scale_updates_gpu, l.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_pull_array(l.bias_updates_gpu, l.bias_updates, l.outputs);
cuda_pull_array(l.weight_updates_gpu, l.weight_updates, l.outputs*l.inputs);
}
}
void push_updates(layer l)
{
if(l.type == CONVOLUTIONAL){
cuda_push_array(l.bias_updates_gpu, l.bias_updates, l.n);
cuda_push_array(l.weight_updates_gpu, l.weight_updates, l.nweights);
if(l.scale_updates) cuda_push_array(l.scale_updates_gpu, l.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_push_array(l.bias_updates_gpu, l.bias_updates, l.outputs);
cuda_push_array(l.weight_updates_gpu, l.weight_updates, l.outputs*l.inputs);
}
}
void update_layer(layer l, network net)
{
int update_batch = net.batch*net.subdivisions;
float rate = get_current_rate(net);
l.t = get_current_batch(net);
if(l.update_gpu){
l.update_gpu(l, update_batch, rate*l.learning_rate_scale, net.momentum, net.decay);
}
}
void merge_updates(layer l, layer base)
{
if (l.type == CONVOLUTIONAL) {
axpy_cpu(l.n, 1, l.bias_updates, 1, base.bias_updates, 1);
axpy_cpu(l.nweights, 1, l.weight_updates, 1, base.weight_updates, 1);
if (l.scale_updates) {
axpy_cpu(l.n, 1, l.scale_updates, 1, base.scale_updates, 1);
}
} else if(l.type == CONNECTED) {
axpy_cpu(l.outputs, 1, l.bias_updates, 1, base.bias_updates, 1);
axpy_cpu(l.outputs*l.inputs, 1, l.weight_updates, 1, base.weight_updates, 1);
}
}
void distribute_updates(layer l, layer base)
{
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
cuda_push_array(l.bias_updates_gpu, base.bias_updates, l.n);
cuda_push_array(l.weight_updates_gpu, base.weight_updates, l.nweights);
if(base.scale_updates) cuda_push_array(l.scale_updates_gpu, base.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_push_array(l.bias_updates_gpu, base.bias_updates, l.outputs);
cuda_push_array(l.weight_updates_gpu, base.weight_updates, l.outputs*l.inputs);
}
}
*/
/*
void sync_layer(network *nets, int n, int j)
{
int i;
network net = nets[0];
layer base = net.layers[j];
scale_weights(base, 0);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i].gpu_index);
layer l = nets[i].layers[j];
pull_weights(l);
merge_weights(l, base);
}
scale_weights(base, 1./n);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i].gpu_index);
layer l = nets[i].layers[j];
distribute_weights(l, base);
}
}
*/
void sync_layer(network **nets, int n, int j)
{
int i;
network *net = nets[0];
layer base = net->layers[j];
scale_weights(base, 0);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i]->gpu_index);
layer l = nets[i]->layers[j];
pull_weights(l);
merge_weights(l, base);
}
scale_weights(base, 1./n);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i]->gpu_index);
layer l = nets[i]->layers[j];
distribute_weights(l, base);
}
}
typedef struct{
network **nets;
int n;
int j;
} sync_args;
void *sync_layer_thread(void *ptr)
{
sync_args args = *(sync_args*)ptr;
sync_layer(args.nets, args.n, args.j);
free(ptr);
return 0;
}
pthread_t sync_layer_in_thread(network **nets, int n, int j)
{
pthread_t thread;
sync_args *ptr = (sync_args *)calloc(1, sizeof(sync_args));
ptr->nets = nets;
ptr->n = n;
ptr->j = j;
if(pthread_create(&thread, 0, sync_layer_thread, ptr)) error("Thread creation failed");
return thread;
}
void sync_nets(network **nets, int n, int interval)
{
int j;
int layers = nets[0]->n;
pthread_t *threads = (pthread_t *) calloc(layers, sizeof(pthread_t));
*(nets[0]->seen) += interval * (n-1) * nets[0]->batch * nets[0]->subdivisions;
for (j = 0; j < n; ++j){
*(nets[j]->seen) = *(nets[0]->seen);
}
for (j = 0; j < layers; ++j) {
threads[j] = sync_layer_in_thread(nets, n, j);
}
for (j = 0; j < layers; ++j) {
pthread_join(threads[j], 0);
}
free(threads);
}
float train_networks(network **nets, int n, data d, int interval)
{
int i;
int batch = nets[0]->batch;
int subdivisions = nets[0]->subdivisions;
assert(batch * subdivisions * n == d.X.rows);
pthread_t *threads = (pthread_t *) calloc(n, sizeof(pthread_t));
float *errors = (float *) calloc(n, sizeof(float));
float sum = 0;
for(i = 0; i < n; ++i){
data p = get_data_part(d, i, n);
threads[i] = train_network_in_thread(nets[i], p, errors + i);
}
for(i = 0; i < n; ++i){
pthread_join(threads[i], 0);
//printf("%f\n", errors[i]);
sum += errors[i];
}
//cudaDeviceSynchronize();
if (get_current_batch(nets[0]) % interval == 0) {
printf("Syncing... ");
fflush(stdout);
sync_nets(nets, n, interval);
printf("Done!\n");
}
//cudaDeviceSynchronize();
free(threads);
free(errors);
return (float)sum/(n);
}
void pull_network_output(network *net)
{
layer l = get_network_output_layer(net);
cuda_pull_array(l.output_gpu, l.output, l.outputs*l.batch);
}
#endif

16
src/network.h
View File

@ -10,22 +10,20 @@
#ifdef GPU
float train_network_datum_gpu(network net);
float *network_predict_gpu(network net, float *input);
void pull_network_output(network net);
void pull_network_output(network *net);
#endif
void compare_networks(network n1, network n2, data d);
void compare_networks(network *n1, network *n2, data d);
char *get_layer_string(LAYER_TYPE a);
network make_network(int n);
network *make_network(int n);
float network_accuracy_multi(network net, data d, int n);
int get_predicted_class_network(network net);
void print_network(network net);
float network_accuracy_multi(network *net, data d, int n);
int get_predicted_class_network(network *net);
void print_network(network *net);
int resize_network(network *net, int w, int h);
void calc_network_cost(network net);
void calc_network_cost(network *net);
#endif

422
src/network_kernels.cu
View File

@ -1,422 +0,0 @@
#include "cuda_runtime.h"
#include "curand.h"
#include "cublas_v2.h"
extern "C" {
#include <stdio.h>
#include <time.h>
#include <assert.h>
#include "network.h"
#include "data.h"
#include "utils.h"
#include "parser.h"
#include "crop_layer.h"
#include "connected_layer.h"
#include "rnn_layer.h"
#include "gru_layer.h"
#include "crnn_layer.h"
#include "detection_layer.h"
#include "region_layer.h"
#include "convolutional_layer.h"
#include "activation_layer.h"
#include "maxpool_layer.h"
#include "reorg_layer.h"
#include "avgpool_layer.h"
#include "normalization_layer.h"
#include "batchnorm_layer.h"
#include "cost_layer.h"
#include "local_layer.h"
#include "softmax_layer.h"
#include "dropout_layer.h"
#include "route_layer.h"
#include "shortcut_layer.h"
#include "blas.h"
}
void forward_network_gpu(network net)
{
int i;
for(i = 0; i < net.n; ++i){
net.index = i;
layer l = net.layers[i];
if(l.delta_gpu){
fill_gpu(l.outputs * l.batch, 0, l.delta_gpu, 1);
}
l.forward_gpu(l, net);
net.input_gpu = l.output_gpu;
net.input = l.output;
if(l.truth) {
net.truth_gpu = l.output_gpu;
net.truth = l.output;
}
}
pull_network_output(net);
calc_network_cost(net);
}
void backward_network_gpu(network net)
{
int i;
network orig = net;
for(i = net.n-1; i >= 0; --i){
layer l = net.layers[i];
if(l.stopbackward) break;
if(i == 0){
net = orig;
}else{
layer prev = net.layers[i-1];
net.input = prev.output;
net.delta = prev.delta;
net.input_gpu = prev.output_gpu;
net.delta_gpu = prev.delta_gpu;
}
net.index = i;
l.backward_gpu(l, net);
}
}
void update_network_gpu(network net)
{
cuda_set_device(net.gpu_index);
int i;
update_args a = {0};
a.batch = net.batch*net.subdivisions;
a.learning_rate = get_current_rate(net);
a.momentum = net.momentum;
a.decay = net.decay;
a.adam = net.adam;
a.B1 = net.B1;
a.B2 = net.B2;
a.eps = net.eps;
++*net.t;
a.t = (*net.t);
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
if(l.update_gpu){
l.update_gpu(l, a);
}
}
}
void harmless_update_network_gpu(network net)
{
cuda_set_device(net.gpu_index);
int i;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
if(l.weight_updates_gpu) fill_gpu(l.nweights, 0, l.weight_updates_gpu, 1);
if(l.bias_updates_gpu) fill_gpu(l.nbiases, 0, l.bias_updates_gpu, 1);
if(l.scale_updates_gpu) fill_gpu(l.nbiases, 0, l.scale_updates_gpu, 1);
}
}
float train_network_datum_gpu(network net)
{
*net.seen += net.batch;
int x_size = net.inputs*net.batch;
int y_size = net.truths*net.batch;
cuda_push_array(net.input_gpu, net.input, x_size);
cuda_push_array(net.truth_gpu, net.truth, y_size);
net.train = 1;
forward_network_gpu(net);
backward_network_gpu(net);
float error = *net.cost;
if (((*net.seen) / net.batch) % net.subdivisions == 0) update_network_gpu(net);
return error;
}
typedef struct {
network net;
data d;
float *err;
} train_args;
void *train_thread(void *ptr)
{
train_args args = *(train_args*)ptr;
free(ptr);
cuda_set_device(args.net.gpu_index);
*args.err = train_network(args.net, args.d);
return 0;
}
pthread_t train_network_in_thread(network net, data d, float *err)
{
pthread_t thread;
train_args *ptr = (train_args *)calloc(1, sizeof(train_args));
ptr->net = net;
ptr->d = d;
ptr->err = err;
if(pthread_create(&thread, 0, train_thread, ptr)) error("Thread creation failed");
return thread;
}
void merge_weights(layer l, layer base)
{
if (l.type == CONVOLUTIONAL) {
axpy_cpu(l.n, 1, l.bias_updates, 1, base.biases, 1);
axpy_cpu(l.nweights, 1, l.weight_updates, 1, base.weights, 1);
if (l.scales) {
axpy_cpu(l.n, 1, l.scale_updates, 1, base.scales, 1);
}
} else if(l.type == CONNECTED) {
axpy_cpu(l.outputs, 1, l.bias_updates, 1, base.biases, 1);
axpy_cpu(l.outputs*l.inputs, 1, l.weight_updates, 1, base.weights, 1);
}
}
void scale_weights(layer l, float s)
{
if (l.type == CONVOLUTIONAL) {
scal_cpu(l.n, s, l.biases, 1);
scal_cpu(l.nweights, s, l.weights, 1);
if (l.scales) {
scal_cpu(l.n, s, l.scales, 1);
}
} else if(l.type == CONNECTED) {
scal_cpu(l.outputs, s, l.biases, 1);
scal_cpu(l.outputs*l.inputs, s, l.weights, 1);
}
}
void pull_weights(layer l)
{
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
cuda_pull_array(l.biases_gpu, l.bias_updates, l.n);
cuda_pull_array(l.weights_gpu, l.weight_updates, l.nweights);
if(l.scales) cuda_pull_array(l.scales_gpu, l.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_pull_array(l.biases_gpu, l.bias_updates, l.outputs);
cuda_pull_array(l.weights_gpu, l.weight_updates, l.outputs*l.inputs);
}
}
void push_weights(layer l)
{
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
cuda_push_array(l.biases_gpu, l.biases, l.n);
cuda_push_array(l.weights_gpu, l.weights, l.nweights);
if(l.scales) cuda_push_array(l.scales_gpu, l.scales, l.n);
} else if(l.type == CONNECTED){
cuda_push_array(l.biases_gpu, l.biases, l.outputs);
cuda_push_array(l.weights_gpu, l.weights, l.outputs*l.inputs);
}
}
void distribute_weights(layer l, layer base)
{
if (l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL) {
cuda_push_array(l.biases_gpu, base.biases, l.n);
cuda_push_array(l.weights_gpu, base.weights, l.nweights);
if (base.scales) cuda_push_array(l.scales_gpu, base.scales, l.n);
} else if (l.type == CONNECTED) {
cuda_push_array(l.biases_gpu, base.biases, l.outputs);
cuda_push_array(l.weights_gpu, base.weights, l.outputs*l.inputs);
}
}
/*
void pull_updates(layer l)
{
if(l.type == CONVOLUTIONAL){
cuda_pull_array(l.bias_updates_gpu, l.bias_updates, l.n);
cuda_pull_array(l.weight_updates_gpu, l.weight_updates, l.nweights);
if(l.scale_updates) cuda_pull_array(l.scale_updates_gpu, l.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_pull_array(l.bias_updates_gpu, l.bias_updates, l.outputs);
cuda_pull_array(l.weight_updates_gpu, l.weight_updates, l.outputs*l.inputs);
}
}
void push_updates(layer l)
{
if(l.type == CONVOLUTIONAL){
cuda_push_array(l.bias_updates_gpu, l.bias_updates, l.n);
cuda_push_array(l.weight_updates_gpu, l.weight_updates, l.nweights);
if(l.scale_updates) cuda_push_array(l.scale_updates_gpu, l.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_push_array(l.bias_updates_gpu, l.bias_updates, l.outputs);
cuda_push_array(l.weight_updates_gpu, l.weight_updates, l.outputs*l.inputs);
}
}
void update_layer(layer l, network net)
{
int update_batch = net.batch*net.subdivisions;
float rate = get_current_rate(net);
l.t = get_current_batch(net);
if(l.update_gpu){
l.update_gpu(l, update_batch, rate*l.learning_rate_scale, net.momentum, net.decay);
}
}
void merge_updates(layer l, layer base)
{
if (l.type == CONVOLUTIONAL) {
axpy_cpu(l.n, 1, l.bias_updates, 1, base.bias_updates, 1);
axpy_cpu(l.nweights, 1, l.weight_updates, 1, base.weight_updates, 1);
if (l.scale_updates) {
axpy_cpu(l.n, 1, l.scale_updates, 1, base.scale_updates, 1);
}
} else if(l.type == CONNECTED) {
axpy_cpu(l.outputs, 1, l.bias_updates, 1, base.bias_updates, 1);
axpy_cpu(l.outputs*l.inputs, 1, l.weight_updates, 1, base.weight_updates, 1);
}
}
void distribute_updates(layer l, layer base)
{
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
cuda_push_array(l.bias_updates_gpu, base.bias_updates, l.n);
cuda_push_array(l.weight_updates_gpu, base.weight_updates, l.nweights);
if(base.scale_updates) cuda_push_array(l.scale_updates_gpu, base.scale_updates, l.n);
} else if(l.type == CONNECTED){
cuda_push_array(l.bias_updates_gpu, base.bias_updates, l.outputs);
cuda_push_array(l.weight_updates_gpu, base.weight_updates, l.outputs*l.inputs);
}
}
*/
/*
void sync_layer(network *nets, int n, int j)
{
int i;
network net = nets[0];
layer base = net.layers[j];
scale_weights(base, 0);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i].gpu_index);
layer l = nets[i].layers[j];
pull_weights(l);
merge_weights(l, base);
}
scale_weights(base, 1./n);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i].gpu_index);
layer l = nets[i].layers[j];
distribute_weights(l, base);
}
}
*/
void sync_layer(network *nets, int n, int j)
{
int i;
network net = nets[0];
layer base = net.layers[j];
scale_weights(base, 0);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i].gpu_index);
layer l = nets[i].layers[j];
pull_weights(l);
merge_weights(l, base);
}
scale_weights(base, 1./n);
for (i = 0; i < n; ++i) {
cuda_set_device(nets[i].gpu_index);
layer l = nets[i].layers[j];
distribute_weights(l, base);
}
}
typedef struct{
network *nets;
int n;
int j;
} sync_args;
void *sync_layer_thread(void *ptr)
{
sync_args args = *(sync_args*)ptr;
sync_layer(args.nets, args.n, args.j);
free(ptr);
return 0;
}
pthread_t sync_layer_in_thread(network *nets, int n, int j)
{
pthread_t thread;
sync_args *ptr = (sync_args *)calloc(1, sizeof(sync_args));
ptr->nets = nets;
ptr->n = n;
ptr->j = j;
if(pthread_create(&thread, 0, sync_layer_thread, ptr)) error("Thread creation failed");
return thread;
}
void sync_nets(network *nets, int n, int interval)
{
int j;
int layers = nets[0].n;
pthread_t *threads = (pthread_t *) calloc(layers, sizeof(pthread_t));
*nets[0].seen += interval * (n-1) * nets[0].batch * nets[0].subdivisions;
for (j = 0; j < n; ++j){
*nets[j].seen = *nets[0].seen;
}
for (j = 0; j < layers; ++j) {
threads[j] = sync_layer_in_thread(nets, n, j);
}
for (j = 0; j < layers; ++j) {
pthread_join(threads[j], 0);
}
free(threads);
}
float train_networks(network *nets, int n, data d, int interval)
{
int i;
int batch = nets[0].batch;
int subdivisions = nets[0].subdivisions;
assert(batch * subdivisions * n == d.X.rows);
pthread_t *threads = (pthread_t *) calloc(n, sizeof(pthread_t));
float *errors = (float *) calloc(n, sizeof(float));
float sum = 0;
for(i = 0; i < n; ++i){
data p = get_data_part(d, i, n);
threads[i] = train_network_in_thread(nets[i], p, errors + i);
}
for(i = 0; i < n; ++i){
pthread_join(threads[i], 0);
//printf("%f\n", errors[i]);
sum += errors[i];
}
//cudaDeviceSynchronize();
if (get_current_batch(nets[0]) % interval == 0) {
printf("Syncing... ");
fflush(stdout);
sync_nets(nets, n, interval);
printf("Done!\n");
}
//cudaDeviceSynchronize();
free(threads);
free(errors);
return (float)sum/(n);
}
void pull_network_output(network net)
{
layer l = get_network_output_layer(net);
cuda_pull_array(l.output_gpu, l.output, l.outputs*l.batch);
}
float *network_predict_gpu(network net, float *input)
{
cuda_set_device(net.gpu_index);
cuda_push_array(net.input_gpu, input, net.inputs*net.batch);
net.truth = 0;
net.train = 0;
forward_network_gpu(net);
return net.output;
}

101
src/parser.c
View File

@ -116,7 +116,7 @@ typedef struct size_params{
int c;
int index;
int time_steps;
network net;
network *net;
} size_params;
local_layer parse_local(list *options, size_params params)
@ -160,7 +160,7 @@ layer parse_deconvolutional(list *options, size_params params)
int padding = option_find_int_quiet(options, "padding",0);
if(pad) padding = size/2;
layer l = make_deconvolutional_layer(batch,h,w,c,n,size,stride,padding, activation, batch_normalize, params.net.adam);
layer l = make_deconvolutional_layer(batch,h,w,c,n,size,stride,padding, activation, batch_normalize, params.net->adam);
return l;
}
@ -189,7 +189,7 @@ convolutional_layer parse_convolutional(list *options, size_params params)
int binary = option_find_int_quiet(options, "binary", 0);
int xnor = option_find_int_quiet(options, "xnor", 0);
convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,groups,size,stride,padding,activation, batch_normalize, binary, xnor, params.net.adam);
convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,groups,size,stride,padding,activation, batch_normalize, binary, xnor, params.net->adam);
layer.flipped = option_find_int_quiet(options, "flipped", 0);
layer.dot = option_find_float_quiet(options, "dot", 0);
@ -218,7 +218,7 @@ layer parse_rnn(list *options, size_params params)
ACTIVATION activation = get_activation(activation_s);
int batch_normalize = option_find_int_quiet(options, "batch_normalize", 0);
layer l = make_rnn_layer(params.batch, params.inputs, output, params.time_steps, activation, batch_normalize, params.net.adam);
layer l = make_rnn_layer(params.batch, params.inputs, output, params.time_steps, activation, batch_normalize, params.net->adam);
l.shortcut = option_find_int_quiet(options, "shortcut", 0);
@ -230,7 +230,7 @@ layer parse_gru(list *options, size_params params)
int output = option_find_int(options, "output",1);
int batch_normalize = option_find_int_quiet(options, "batch_normalize", 0);
layer l = make_gru_layer(params.batch, params.inputs, output, params.time_steps, batch_normalize, params.net.adam);
layer l = make_gru_layer(params.batch, params.inputs, output, params.time_steps, batch_normalize, params.net->adam);
l.tanh = option_find_int_quiet(options, "tanh", 0);
return l;
@ -241,7 +241,7 @@ layer parse_lstm(list *options, size_params params)
int output = option_find_int(options, "output", 1);
int batch_normalize = option_find_int_quiet(options, "batch_normalize", 0);
layer l = make_lstm_layer(params.batch, params.inputs, output, params.time_steps, batch_normalize, params.net.adam);
layer l = make_lstm_layer(params.batch, params.inputs, output, params.time_steps, batch_normalize, params.net->adam);
return l;
}
@ -253,7 +253,7 @@ layer parse_connected(list *options, size_params params)
ACTIVATION activation = get_activation(activation_s);
int batch_normalize = option_find_int_quiet(options, "batch_normalize", 0);
layer l = make_connected_layer(params.batch, params.inputs, output, activation, batch_normalize, params.net.adam);
layer l = make_connected_layer(params.batch, params.inputs, output, activation, batch_normalize, params.net->adam);
return l;
}
@ -456,14 +456,14 @@ layer parse_batchnorm(list *options, size_params params)
return l;
}
layer parse_shortcut(list *options, size_params params, network net)
layer parse_shortcut(list *options, size_params params, network *net)
{
char *l = option_find(options, "from");
int index = atoi(l);
if(index < 0) index = params.index + index;
int batch = params.batch;
layer from = net.layers[index];
layer from = net->layers[index];
layer s = make_shortcut_layer(batch, index, params.w, params.h, params.c, from.out_w, from.out_h, from.out_c);
@ -491,7 +491,7 @@ layer parse_activation(list *options, size_params params)
return l;
}
route_layer parse_route(list *options, size_params params, network net)
route_layer parse_route(list *options, size_params params, network *net)
{
char *l = option_find(options, "layers");
int len = strlen(l);
@ -509,19 +509,19 @@ route_layer parse_route(list *options, size_params params, network net)
l = strchr(l, ',')+1;
if(index < 0) index = params.index + index;
layers[i] = index;
sizes[i] = net.layers[index].outputs;
sizes[i] = net->layers[index].outputs;
}
int batch = params.batch;
route_layer layer = make_route_layer(batch, n, layers, sizes);
convolutional_layer first = net.layers[layers[0]];
convolutional_layer first = net->layers[layers[0]];
layer.out_w = first.out_w;
layer.out_h = first.out_h;
layer.out_c = first.out_c;
for(i = 1; i < n; ++i){
int index = layers[i];
convolutional_layer next = net.layers[index];
convolutional_layer next = net->layers[index];
if(next.out_w == first.out_w && next.out_h == first.out_h){
layer.out_c += next.out_c;
}else{
@ -557,6 +557,7 @@ void parse_net_options(list *options, network *net)
net->batch /= subdivs;
net->batch *= net->time_steps;
net->subdivisions = subdivs;
net->random = option_find_int_quiet(options, "random", 0);
net->adam = option_find_int_quiet(options, "adam", 0);
if(net->adam){
@ -571,6 +572,8 @@ void parse_net_options(list *options, network *net)
net->inputs = option_find_int_quiet(options, "inputs", net->h * net->w * net->c);
net->max_crop = option_find_int_quiet(options, "max_crop",net->w*2);
net->min_crop = option_find_int_quiet(options, "min_crop",net->w);
net->max_ratio = option_find_float_quiet(options, "max_ratio", (float) net->max_crop / net->w);
net->min_ratio = option_find_float_quiet(options, "min_ratio", (float) net->min_crop / net->w);
net->center = option_find_int_quiet(options, "center",0);
net->angle = option_find_float_quiet(options, "angle", 0);
@ -628,26 +631,26 @@ int is_network(section *s)
|| strcmp(s->type, "[network]")==0);
}
network parse_network_cfg(char *filename)
network *parse_network_cfg(char *filename)
{
list *sections = read_cfg(filename);
node *n = sections->front;
if(!n) error("Config file has no sections");
network net = make_network(sections->size - 1);
net.gpu_index = gpu_index;
network *net = make_network(sections->size - 1);
net->gpu_index = gpu_index;
size_params params;
section *s = (section *)n->val;
list *options = s->options;
if(!is_network(s)) error("First section must be [net] or [network]");
parse_net_options(options, &net);
parse_net_options(options, net);
params.h = net.h;
params.w = net.w;
params.c = net.c;
params.inputs = net.inputs;
params.batch = net.batch;
params.time_steps = net.time_steps;
params.h = net->h;
params.w = net->w;
params.c = net->c;
params.inputs = net->inputs;
params.batch = net->batch;
params.time_steps = net->time_steps;
params.net = net;
size_t workspace_size = 0;
@ -690,7 +693,7 @@ network parse_network_cfg(char *filename)
l = parse_detection(options, params);
}else if(lt == SOFTMAX){
l = parse_softmax(options, params);
net.hierarchy = l.softmax_tree;
net->hierarchy = l.softmax_tree;
}else if(lt == NORMALIZATION){
l = parse_normalization(options, params);
}else if(lt == BATCHNORM){
@ -707,11 +710,11 @@ network parse_network_cfg(char *filename)
l = parse_shortcut(options, params, net);
}else if(lt == DROPOUT){
l = parse_dropout(options, params);
l.output = net.layers[count-1].output;
l.delta = net.layers[count-1].delta;
l.output = net->layers[count-1].output;
l.delta = net->layers[count-1].delta;
#ifdef GPU
l.output_gpu = net.layers[count-1].output_gpu;
l.delta_gpu = net.layers[count-1].delta_gpu;
l.output_gpu = net->layers[count-1].output_gpu;
l.delta_gpu = net->layers[count-1].delta_gpu;
#endif
}else{
fprintf(stderr, "Type not recognized: %s\n", s->type);
@ -724,7 +727,7 @@ network parse_network_cfg(char *filename)
l.learning_rate_scale = option_find_float_quiet(options, "learning_rate", 1);
l.smooth = option_find_float_quiet(options, "smooth", 0);
option_unused(options);
net.layers[count] = l;
net->layers[count] = l;
if (l.workspace_size > workspace_size) workspace_size = l.workspace_size;
free_section(s);
n = n->next;
@ -738,27 +741,27 @@ network parse_network_cfg(char *filename)
}
free_list(sections);
layer out = get_network_output_layer(net);
net.outputs = out.outputs;
net.truths = out.outputs;
if(net.layers[net.n-1].truths) net.truths = net.layers[net.n-1].truths;
net.output = out.output;
net.input = calloc(net.inputs*net.batch, sizeof(float));
net.truth = calloc(net.truths*net.batch, sizeof(float));
net->outputs = out.outputs;
net->truths = out.outputs;
if(net->layers[net->n-1].truths) net->truths = net->layers[net->n-1].truths;
net->output = out.output;
net->input = calloc(net->inputs*net->batch, sizeof(float));
net->truth = calloc(net->truths*net->batch, sizeof(float));
#ifdef GPU
net.output_gpu = out.output_gpu;
net.input_gpu = cuda_make_array(net.input, net.inputs*net.batch);
net.truth_gpu = cuda_make_array(net.truth, net.truths*net.batch);
net->output_gpu = out.output_gpu;
net->input_gpu = cuda_make_array(net->input, net->inputs*net->batch);
net->truth_gpu = cuda_make_array(net->truth, net->truths*net->batch);
#endif
if(workspace_size){
//printf("%ld\n", workspace_size);
#ifdef GPU
if(gpu_index >= 0){
net.workspace = cuda_make_array(0, (workspace_size-1)/sizeof(float)+1);
net->workspace = cuda_make_array(0, (workspace_size-1)/sizeof(float)+1);
}else {
net.workspace = calloc(1, workspace_size);
net->workspace = calloc(1, workspace_size);
}
#else
net.workspace = calloc(1, workspace_size);
net->workspace = calloc(1, workspace_size);
#endif
}
return net;
@ -880,11 +883,11 @@ void save_connected_weights(layer l, FILE *fp)
}
}
void save_weights_upto(network net, char *filename, int cutoff)
void save_weights_upto(network *net, char *filename, int cutoff)
{
#ifdef GPU
if(net.gpu_index >= 0){
cuda_set_device(net.gpu_index);
if(net->gpu_index >= 0){
cuda_set_device(net->gpu_index);
}
#endif
fprintf(stderr, "Saving weights to %s\n", filename);
@ -897,11 +900,11 @@ void save_weights_upto(network net, char *filename, int cutoff)
fwrite(&major, sizeof(int), 1, fp);
fwrite(&minor, sizeof(int), 1, fp);
fwrite(&revision, sizeof(int), 1, fp);
fwrite(net.seen, sizeof(size_t), 1, fp);
fwrite(net->seen, sizeof(size_t), 1, fp);
int i;
for(i = 0; i < net.n && i < cutoff; ++i){
layer l = net.layers[i];
for(i = 0; i < net->n && i < cutoff; ++i){
layer l = net->layers[i];
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
save_convolutional_weights(l, fp);
} if(l.type == CONNECTED){
@ -952,9 +955,9 @@ void save_weights_upto(network net, char *filename, int cutoff)
}
fclose(fp);
}
void save_weights(network net, char *filename)
void save_weights(network *net, char *filename)
{
save_weights_upto(net, filename, net.n);
save_weights_upto(net, filename, net->n);
}
void transpose_matrix(float *a, int rows, int cols)

8
src/region_layer.c
View File

@ -109,7 +109,7 @@ void delta_region_mask(float *truth, float *x, int n, int index, float *delta, i
}
void delta_region_class(float *output, float *delta, int index, int class, int classes, tree *hier, float scale, int stride, float *avg_cat)
void delta_region_class(float *output, float *delta, int index, int class, int classes, tree *hier, float scale, int stride, float *avg_cat, int tag)
{
int i, n;
if(hier){
@ -127,7 +127,7 @@ void delta_region_class(float *output, float *delta, int index, int class, int c
}
*avg_cat += pred;
} else {
if (delta[index]){
if (delta[index] && tag){
delta[index + stride*class] = scale * (1 - output[index + stride*class]);
return;
}
@ -218,7 +218,7 @@ void forward_region_layer(const layer l, network net)
}
int class_index = entry_index(l, b, maxi, l.coords + 1);
int obj_index = entry_index(l, b, maxi, l.coords);
delta_region_class(l.output, l.delta, class_index, class, l.classes, l.softmax_tree, l.class_scale, l.w*l.h, &avg_cat);
delta_region_class(l.output, l.delta, class_index, class, l.classes, l.softmax_tree, l.class_scale, l.w*l.h, &avg_cat, !l.softmax);
if(l.output[obj_index] < .3) l.delta[obj_index] = l.object_scale * (.3 - l.output[obj_index]);
else l.delta[obj_index] = 0;
l.delta[obj_index] = 0;
@ -316,7 +316,7 @@ void forward_region_layer(const layer l, network net)
int class = net.truth[t*(l.coords + 1) + b*l.truths + l.coords];
if (l.map) class = l.map[class];
int class_index = entry_index(l, b, best_n*l.w*l.h + j*l.w + i, l.coords + 1);
delta_region_class(l.output, l.delta, class_index, class, l.classes, l.softmax_tree, l.class_scale, l.w*l.h, &avg_cat);
delta_region_class(l.output, l.delta, class_index, class, l.classes, l.softmax_tree, l.class_scale, l.w*l.h, &avg_cat, !l.softmax);
++count;
++class_count;
}