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;
}
void save_weights_upto(network net, char *filename, int cutoff)
{
#ifdef GPU
if(net.gpu_index >= 0){
cuda_set_device(net.gpu_index);
}
#endif
fprintf(stderr, "Saving weights to %s\n", filename);
FILE *fp = fopen(filename, "wb");
if(!fp) file_error(filename);
int major = 0;
int minor = 1;
int revision = 0;
fwrite(&major, sizeof(int), 1, fp);
fwrite(&minor, sizeof(int), 1, fp);
fwrite(&revision, sizeof(int), 1, fp);
fwrite(net.seen, sizeof(int), 1, fp);
int i;
for(i = 0; i < net.n && i < cutoff; ++i){
layer l = net.layers[i];
if(l.type == CONVOLUTIONAL){
save_convolutional_weights(l, fp);
} if(l.type == CONNECTED){
save_connected_weights(l, fp);
} if(l.type == BATCHNORM){
save_batchnorm_weights(l, fp);
} if(l.type == RNN){
save_connected_weights(*(l.input_layer), fp);
save_connected_weights(*(l.self_layer), fp);
save_connected_weights(*(l.output_layer), fp);
} if(l.type == GRU){
save_connected_weights(*(l.input_z_layer), fp);
save_connected_weights(*(l.input_r_layer), fp);
save_connected_weights(*(l.input_h_layer), fp);
save_connected_weights(*(l.state_z_layer), fp);
save_connected_weights(*(l.state_r_layer), fp);
save_connected_weights(*(l.state_h_layer), fp);
} if(l.type == CRNN){
save_convolutional_weights(*(l.input_layer), fp);
save_convolutional_weights(*(l.self_layer), fp);
save_convolutional_weights(*(l.output_layer), fp);
} if(l.type == LOCAL){
#ifdef GPU
if(gpu_index >= 0){
pull_local_layer(l);
}
#endif
int locations = l.out_w*l.out_h;
int size = l.size*l.size*l.c*l.n*locations;
fwrite(l.biases, sizeof(float), l.outputs, fp);
fwrite(l.weights, sizeof(float), size, fp);
}
}
fclose(fp);
}
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;
}
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_ongpu(l.nweights, 0, l.weight_updates_gpu, 1);
if(l.bias_updates_gpu) fill_ongpu(l.nbiases, 0, l.bias_updates_gpu, 1);
if(l.scale_updates_gpu) fill_ongpu(l.nbiases, 0, l.scale_updates_gpu, 1);
}
}
void sync_layer(network *nets, int n, int j)
{
//printf("Syncing layer %d\n", j);
int i;
network net = nets[0];
layer base = net.layers[j];
cuda_set_device(net.gpu_index);
pull_weights(base);
for (i = 1; 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);
}
//printf("Done syncing layer %d\n", j);
}
void update_network_gpu(network net)
{
cuda_set_device(net.gpu_index);
int i;
int update_batch = net.batch*net.subdivisions;
float rate = get_current_rate(net);
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
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 train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, int ngpus, int clear)
{
list *options = read_data_cfg(datacfg);
char *train_images = option_find_str(options, "train", "data/train.list");
char *backup_directory = option_find_str(options, "backup", "/backup/");
srand(time(0));
char *base = basecfg(cfgfile);
printf("%s\n", base);
float avg_loss = -1;
network *nets = calloc(ngpus, sizeof(network));
srand(time(0));
int seed = rand();
int i;
for(i = 0; i < ngpus; ++i){
srand(seed);
#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;
}
srand(time(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);
data train, buffer;
layer l = net.layers[net.n - 1];
int classes = l.classes;
float jitter = l.jitter;
list *plist = get_paths(train_images);
//int N = plist->size;
char **paths = (char **)list_to_array(plist);
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.classes = classes;
args.jitter = jitter;
args.num_boxes = l.max_boxes;
args.d = &buffer;
args.type = DETECTION_DATA;
args.threads = 8;
args.angle = net.angle;
args.exposure = net.exposure;
args.saturation = net.saturation;
args.hue = net.hue;
pthread_t load_thread = load_data(args);
clock_t time;
int count = 0;
//while(i*imgs < N*120){
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;
//int dim = (rand() % 4 + 16) * 32;
int dim = (args.w <= args.h ? args.w : args.h);
printf("%d\n", dim);
args.w = dim;
args.h = dim;
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=clock();
pthread_join(load_thread, 0);
train = buffer;
load_thread = load_data(args);
/*
int k;
for(k = 0; k < l.max_boxes; ++k){
box b = float_to_box(train.y.vals[10] + 1 + k*5);
if(!b.x) break;
printf("loaded: %f %f %f %f\n", b.x, b.y, b.w, b.h);
}
image im = float_to_image(448, 448, 3, train.X.vals[10]);
int k;
for(k = 0; k < l.max_boxes; ++k){
box b = float_to_box(train.y.vals[10] + 1 + k*5);
printf("%d %d %d %d\n", truth.x, truth.y, truth.w, truth.h);
draw_bbox(im, b, 8, 1,0,0);
}
save_image(im, "truth11");
*/
printf("Loaded: %lf seconds\n", sec(clock()-time));
time=clock();
float loss = 0;
#ifdef GPU
if(ngpus == 1){
loss = train_network(net, train);
} else {
loss = train_networks(nets, ngpus, train, 4);
}
#else
loss = train_network(net, train);
#endif
if (avg_loss < 0) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
i = get_current_batch(net);
printf("%d: %f, %f avg, %f rate, %lf seconds, %d images\n", get_current_batch(net), loss, avg_loss, get_current_rate(net), sec(clock()-time), i*imgs);
if(i%1000==0 || (i < 1000 && i%100 == 0)){
#ifdef GPU
if(ngpus != 1) sync_nets(nets, ngpus, 0);
#endif
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
save_weights(net, buff);
}
free_data(train);
}
#ifdef GPU
if(ngpus != 1) sync_nets(nets, ngpus, 0);
#endif
char buff[256];
sprintf(buff, "%s/%s_final.weights", backup_directory, base);
save_weights(net, buff);
}
void load_weights_upto(network *net, char *filename, int cutoff)
{
#ifdef GPU
if(net->gpu_index >= 0){
cuda_set_device(net->gpu_index);
}
#endif
fprintf(stderr, "Loading weights from %s...", filename);
fflush(stdout);
FILE *fp = fopen(filename, "rb");
if(!fp) file_error(filename);
int major;
int minor;
int revision;
fread(&major, sizeof(int), 1, fp);
fread(&minor, sizeof(int), 1, fp);
fread(&revision, sizeof(int), 1, fp);
fread(net->seen, sizeof(int), 1, fp);
int transpose = (major > 1000) || (minor > 1000);
int i;
for(i = 0; i < net->n && i < cutoff; ++i){
layer l = net->layers[i];
if (l.dontload) continue;
if(l.type == CONVOLUTIONAL){
load_convolutional_weights(l, fp);
}
if(l.type == CONNECTED){
load_connected_weights(l, fp, transpose);
}
if(l.type == BATCHNORM){
load_batchnorm_weights(l, fp);
}
if(l.type == CRNN){
load_convolutional_weights(*(l.input_layer), fp);
load_convolutional_weights(*(l.self_layer), fp);
load_convolutional_weights(*(l.output_layer), fp);
}
if(l.type == RNN){
load_connected_weights(*(l.input_layer), fp, transpose);
load_connected_weights(*(l.self_layer), fp, transpose);
load_connected_weights(*(l.output_layer), fp, transpose);
}
if(l.type == GRU){
load_connected_weights(*(l.input_z_layer), fp, transpose);
load_connected_weights(*(l.input_r_layer), fp, transpose);
load_connected_weights(*(l.input_h_layer), fp, transpose);
load_connected_weights(*(l.state_z_layer), fp, transpose);
load_connected_weights(*(l.state_r_layer), fp, transpose);
load_connected_weights(*(l.state_h_layer), fp, transpose);
}
if(l.type == LOCAL){
int locations = l.out_w*l.out_h;
int size = l.size*l.size*l.c*l.n*locations;
fread(l.biases, sizeof(float), l.outputs, fp);
fread(l.weights, sizeof(float), size, fp);
#ifdef GPU
if(gpu_index >= 0){
push_local_layer(l);
}
#endif
}
}
fprintf(stderr, "Done!\n");
fclose(fp);
}
int resize_network(network *net, int w, int h)
{
#ifdef GPU
cuda_set_device(net->gpu_index);
cuda_free(net->workspace);
#endif
int i;
//if(w == net->w && h == net->h) return 0;
net->w = w;
net->h = h;
int inputs = 0;
size_t workspace_size = 0;
//fprintf(stderr, "Resizing to %d x %d...\n", w, h);
//fflush(stderr);
for (i = 0; i < net->n; ++i){
layer l = net->layers[i];
if(l.type == CONVOLUTIONAL){
resize_convolutional_layer(&l, w, h);
}else if(l.type == CROP){
resize_crop_layer(&l, w, h);
}else if(l.type == MAXPOOL){
resize_maxpool_layer(&l, w, h);
}else if(l.type == REGION){
resize_region_layer(&l, w, h);
}else if(l.type == ROUTE){
resize_route_layer(&l, net);
}else if(l.type == REORG){
resize_reorg_layer(&l, w, h);
}else if(l.type == AVGPOOL){
resize_avgpool_layer(&l, w, h);
}else if(l.type == NORMALIZATION){
resize_normalization_layer(&l, w, h);
}else if(l.type == COST){
resize_cost_layer(&l, inputs);
}else{
error("Cannot resize this type of layer");
}
if(l.workspace_size > workspace_size) workspace_size = l.workspace_size;
inputs = l.outputs;
net->layers[i] = l;
w = l.out_w;
h = l.out_h;
if(l.type == AVGPOOL) break;
}
layer out = get_network_output_layer(*net);
net->inputs = net->layers[0].inputs;
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;
free(net->input);
free(net->truth);
net->input = calloc(net->inputs*net->batch, sizeof(float));
net->truth = calloc(net->truths*net->batch, sizeof(float));
#ifdef GPU
if(gpu_index >= 0){
cuda_free(net->input_gpu);
cuda_free(net->truth_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->workspace = cuda_make_array(0, (workspace_size-1)/sizeof(float)+1);
}else {
free(net->workspace);
net->workspace = calloc(1, workspace_size);
}
#else
free(net->workspace);
net->workspace = calloc(1, workspace_size);
#endif
//fprintf(stderr, " Done!\n");
return 0;
}
void train_regressor(char *datacfg, char *cfgfile, char *weightfile, int *gpus, int ngpus, int clear)
{
int i;
float avg_loss = -1;
char *base = basecfg(cfgfile);
printf("%s\n", base);
printf("%d\n", ngpus);
network **nets = calloc(ngpus, sizeof(network*));
srand(time(0));
int seed = rand();
for(i = 0; i < ngpus; ++i){
srand(seed);
#ifdef GPU
cuda_set_device(gpus[i]);
#endif
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i]->learning_rate *= ngpus;
}
srand(time(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);
list *options = read_data_cfg(datacfg);
char *backup_directory = option_find_str(options, "backup", "/backup/");
char *train_list = option_find_str(options, "train", "data/train.list");
int classes = option_find_int(options, "classes", 1);
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 = 32;
args.classes = classes;
args.min = net->min_ratio*net->w;
args.max = net->max_ratio*net->w;
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;
args.m = N;
args.type = REGRESSION_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();
float loss = 0;
#ifdef GPU
if(ngpus == 1){
loss = train_network(net, train);
} else {
loss = train_networks(nets, ngpus, train, 4);
}
#else
loss = train_network(net, train);
#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);
free_data(train);
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);
for(i = 0; i < ngpus; ++i){
free_network(nets[i]);
}
free(nets);
free_ptrs((void**)paths, plist->size);
free_list(plist);
free(base);
}
void save_weights_upto(network *net, char *filename, int cutoff)
{
#ifdef GPU
if(net->gpu_index >= 0){
cuda_set_device(net->gpu_index);
}
#endif
fprintf(stderr, "Saving weights to %s\n", filename);
FILE *fp = fopen(filename, "wb");
if(!fp) file_error(filename);
int major = 0;
int minor = 2;
int revision = 0;
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);
int i;
for(i = 0; i < net->n && i < cutoff; ++i){
layer l = net->layers[i];
if (l.dontsave) continue;
if(l.type == CONVOLUTIONAL || l.type == DECONVOLUTIONAL){
save_convolutional_weights(l, fp);
} if(l.type == CONNECTED){
save_connected_weights(l, fp);
} if(l.type == BATCHNORM){
save_batchnorm_weights(l, fp);
} if(l.type == RNN){
save_connected_weights(*(l.input_layer), fp);
save_connected_weights(*(l.self_layer), fp);
save_connected_weights(*(l.output_layer), fp);
} if (l.type == LSTM) {
save_connected_weights(*(l.wi), fp);
save_connected_weights(*(l.wf), fp);
save_connected_weights(*(l.wo), fp);
save_connected_weights(*(l.wg), fp);
save_connected_weights(*(l.ui), fp);
save_connected_weights(*(l.uf), fp);
save_connected_weights(*(l.uo), fp);
save_connected_weights(*(l.ug), fp);
} if (l.type == GRU) {
if(1){
save_connected_weights(*(l.wz), fp);
save_connected_weights(*(l.wr), fp);
save_connected_weights(*(l.wh), fp);
save_connected_weights(*(l.uz), fp);
save_connected_weights(*(l.ur), fp);
save_connected_weights(*(l.uh), fp);
}else{
save_connected_weights(*(l.reset_layer), fp);
save_connected_weights(*(l.update_layer), fp);
save_connected_weights(*(l.state_layer), fp);
}
} if(l.type == CRNN){
save_convolutional_weights(*(l.input_layer), fp);
save_convolutional_weights(*(l.self_layer), fp);
save_convolutional_weights(*(l.output_layer), fp);
} if(l.type == LOCAL){
#ifdef GPU
if(gpu_index >= 0){
pull_local_layer(l);
}
#endif
int locations = l.out_w*l.out_h;
int size = l.size*l.size*l.c*l.n*locations;
fwrite(l.biases, sizeof(float), l.outputs, fp);
fwrite(l.weights, sizeof(float), size, fp);
}
}
fclose(fp);
}
/** returns 1 if and only if the GPU with the given id is usable for
CUDA computations. Only devices with compute capability of 1.1 or
higher are ok, since atomic operations are required for
CUDA-LB. */
int tclcommand_cuda(ClientData data, Tcl_Interp *interp,
int argc, char **argv)
{
#ifndef CUDA
Tcl_AppendResult(interp, "Feature CUDA required!", (char *)NULL);
return TCL_ERROR;
#else
if (argc <= 1) {
Tcl_AppendResult(interp, "too few arguments to the cuda command", (char *)NULL);
return TCL_ERROR;
}
argc--; argv++;
if (ARG0_IS_S("list")) {
if (argc != 1) {
Tcl_AppendResult(interp, "cuda list takes no arguments", (char *)NULL);
return TCL_ERROR;
}
return list_gpus(interp);
}
else if (ARG0_IS_S("setdevice")) {
int dev;
if (argc <= 1 || !ARG1_IS_I(dev)) {
Tcl_AppendResult(interp, "expected: cuda setdevice <devnr>", (char *)NULL);
return TCL_ERROR;
}
if (cuda_check_gpu(dev) == ES_ERROR) {
Tcl_AppendResult(interp, "GPU not present or compute model not sufficient", (char *)NULL);
return TCL_ERROR;
}
if (cuda_set_device(dev) == ES_OK) {
return TCL_OK;
}
else {
Tcl_AppendResult(interp, cuda_error, (char *)NULL);
return TCL_ERROR;
}
}
else if (ARG0_IS_S("getdevice")) {
if (argc != 1) {
Tcl_AppendResult(interp, "cuda getdevice takes no arguments", (char *)NULL);
return TCL_ERROR;
}
int dev = cuda_get_device();
if (dev >= 0) {
char buffer[TCL_INTEGER_SPACE];
sprintf(buffer, "%d", dev);
Tcl_AppendResult(interp, buffer, (char *)NULL);
return TCL_OK;
}
else {
Tcl_AppendResult(interp, cuda_error, (char *)NULL);
return TCL_ERROR;
}
}
else {
Tcl_AppendResult(interp, "unknown subcommand \"", argv[0], "\"", (char *)NULL);
return TCL_ERROR;
}
#endif /* defined(CUDA) */
}
int resize_network(network *net, int w, int h)
{
#ifdef GPU
cuda_set_device(net->gpu_index);
if(gpu_index >= 0){
cuda_free(net->workspace);
if (net->input_gpu) {
cuda_free(*net->input_gpu);
*net->input_gpu = 0;
cuda_free(*net->truth_gpu);
*net->truth_gpu = 0;
}
}
#endif
int i;
//if(w == net->w && h == net->h) return 0;
net->w = w;
net->h = h;
int inputs = 0;
size_t workspace_size = 0;
//fprintf(stderr, "Resizing to %d x %d...\n", w, h);
//fflush(stderr);
for (i = 0; i < net->n; ++i){
layer l = net->layers[i];
//printf(" %d: layer = %d,", i, l.type);
if(l.type == CONVOLUTIONAL){
resize_convolutional_layer(&l, w, h);
}else if(l.type == CROP){
resize_crop_layer(&l, w, h);
}else if(l.type == MAXPOOL){
resize_maxpool_layer(&l, w, h);
}else if(l.type == REGION){
resize_region_layer(&l, w, h);
}else if (l.type == YOLO) {
resize_yolo_layer(&l, w, h);
}else if(l.type == ROUTE){
resize_route_layer(&l, net);
}else if (l.type == SHORTCUT) {
resize_shortcut_layer(&l, w, h);
}else if (l.type == UPSAMPLE) {
resize_upsample_layer(&l, w, h);
}else if(l.type == REORG){
resize_reorg_layer(&l, w, h);
}else if(l.type == AVGPOOL){
resize_avgpool_layer(&l, w, h);
}else if(l.type == NORMALIZATION){
resize_normalization_layer(&l, w, h);
}else if(l.type == COST){
resize_cost_layer(&l, inputs);
}else{
fprintf(stderr, "Resizing type %d \n", (int)l.type);
error("Cannot resize this type of layer");
}
if(l.workspace_size > workspace_size) workspace_size = l.workspace_size;
inputs = l.outputs;
net->layers[i] = l;
w = l.out_w;
h = l.out_h;
if(l.type == AVGPOOL) break;
}
#ifdef GPU
if(gpu_index >= 0){
printf(" try to allocate workspace = %zu * sizeof(float), ", workspace_size / sizeof(float) + 1);
net->workspace = cuda_make_array(0, workspace_size/sizeof(float) + 1);
printf(" CUDA allocate done! \n");
}else {
free(net->workspace);
net->workspace = calloc(1, workspace_size);
}
#else
free(net->workspace);
net->workspace = calloc(1, workspace_size);
#endif
//fprintf(stderr, " Done!\n");
return 0;
}
int main(int argc, char **argv)
{
//test_resize("data/bad.jpg");
//test_box();
//test_convolutional_layer();
if(argc < 2){
fprintf(stderr, "usage: %s <function>\n", argv[0]);
return 0;
}
gpu_index = find_int_arg(argc, argv, "-i", 0);
if(find_arg(argc, argv, "-nogpu")) {
gpu_index = -1;
}
#ifndef GPU
gpu_index = -1;
#else
if(gpu_index >= 0){
cuda_set_device(gpu_index);
}
#endif
if (0 == strcmp(argv[1], "average")){
average(argc, argv);
} else if (0 == strcmp(argv[1], "yolo")){
run_yolo(argc, argv);
} else if (0 == strcmp(argv[1], "super")){
run_super(argc, argv);
} else if (0 == strcmp(argv[1], "lsd")){
run_lsd(argc, argv);
} else if (0 == strcmp(argv[1], "detector")){
run_detector(argc, argv);
} else if (0 == strcmp(argv[1], "detect")){
float thresh = find_float_arg(argc, argv, "-thresh", .5);
char *filename = (argc > 4) ? argv[4]: 0;
char *outfile = find_char_arg(argc, argv, "-out", 0);
int fullscreen = find_arg(argc, argv, "-fullscreen");
test_detector("cfg/coco.data", argv[2], argv[3], filename, thresh, .5, outfile, fullscreen);
} else if (0 == strcmp(argv[1], "cifar")){
run_cifar(argc, argv);
} else if (0 == strcmp(argv[1], "go")){
run_go(argc, argv);
} else if (0 == strcmp(argv[1], "rnn")){
run_char_rnn(argc, argv);
} else if (0 == strcmp(argv[1], "coco")){
run_coco(argc, argv);
} else if (0 == strcmp(argv[1], "classify")){
predict_classifier("cfg/imagenet1k.data", argv[2], argv[3], argv[4], 5);
} else if (0 == strcmp(argv[1], "classifier")){
run_classifier(argc, argv);
} else if (0 == strcmp(argv[1], "regressor")){
run_regressor(argc, argv);
} else if (0 == strcmp(argv[1], "isegmenter")){
run_isegmenter(argc, argv);
} else if (0 == strcmp(argv[1], "segmenter")){
run_segmenter(argc, argv);
} else if (0 == strcmp(argv[1], "art")){
run_art(argc, argv);
} else if (0 == strcmp(argv[1], "tag")){
run_tag(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")){
test_resize(argv[2]);
} else if (0 == strcmp(argv[1], "nightmare")){
run_nightmare(argc, argv);
} else if (0 == strcmp(argv[1], "rgbgr")){
rgbgr_net(argv[2], argv[3], argv[4]);
} else if (0 == strcmp(argv[1], "reset")){
reset_normalize_net(argv[2], argv[3], argv[4]);
} else if (0 == strcmp(argv[1], "denormalize")){
denormalize_net(argv[2], argv[3], argv[4]);
} else if (0 == strcmp(argv[1], "statistics")){
statistics_net(argv[2], argv[3]);
} else if (0 == strcmp(argv[1], "normalize")){
normalize_net(argv[2], argv[3], argv[4]);
} else if (0 == strcmp(argv[1], "rescale")){
rescale_net(argv[2], argv[3], argv[4]);
} else if (0 == strcmp(argv[1], "ops")){
operations(argv[2]);
} else if (0 == strcmp(argv[1], "speed")){
speed(argv[2], (argc > 3 && argv[3]) ? atoi(argv[3]) : 0);
} else if (0 == strcmp(argv[1], "oneoff")){
oneoff(argv[2], argv[3], argv[4]);
} else if (0 == strcmp(argv[1], "oneoff2")){
oneoff2(argv[2], argv[3], argv[4], atoi(argv[5]));
} else if (0 == strcmp(argv[1], "print")){
print_weights(argv[2], argv[3], atoi(argv[4]));
} else if (0 == strcmp(argv[1], "partial")){
partial(argv[2], argv[3], argv[4], atoi(argv[5]));
} else if (0 == strcmp(argv[1], "average")){
average(argc, argv);
} else if (0 == strcmp(argv[1], "visualize")){
visualize(argv[2], (argc > 3) ? argv[3] : 0);
} else if (0 == strcmp(argv[1], "mkimg")){
mkimg(argv[2], argv[3], atoi(argv[4]), atoi(argv[5]), atoi(argv[6]), argv[7]);
} else if (0 == strcmp(argv[1], "imtest")){
test_resize(argv[2]);
} else {
fprintf(stderr, "Not an option: %s\n", argv[1]);
}
return 0;
}
void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus, int ngpus, int clear)
{
int i;
float avg_loss = -1;
char *base = basecfg(cfgfile);
printf("%s\n", base);
printf("%d\n", ngpus);
network **nets = calloc(ngpus, sizeof(network*));
srand(time(0));
int seed = rand();
for(i = 0; i < ngpus; ++i){
srand(seed);
#ifdef GPU
cuda_set_device(gpus[i]);
#endif
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i]->learning_rate *= ngpus;
}
srand(time(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);
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;
double time;
load_args args = {0};
args.w = net->w;
args.h = net->h;
args.threads = 32;
args.hierarchy = net->hierarchy;
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;
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 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);
train = buffer;
load_thread = load_data(args);
printf("Loaded: %lf seconds\n", what_time_is_it_now()-time);
time = what_time_is_it_now();
float loss = 0;
#ifdef GPU
if(ngpus == 1){
loss = train_network(net, train);
} else {
loss = train_networks(nets, ngpus, train, 4);
}
#else
loss = train_network(net, train);
#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);
free_data(train);
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)%1000 == 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);
pthread_join(load_thread, 0);
free_network(net);
free_ptrs((void**)labels, classes);
free_ptrs((void**)paths, plist->size);
free_list(plist);
free(base);
}
void train_segmenter(char *datacfg, char *cfgfile, char *weightfile, int *gpus, int ngpus, int clear, int display)
{
int i;
float avg_loss = -1;
char *base = basecfg(cfgfile);
printf("%s\n", base);
printf("%d\n", ngpus);
network ** nets = calloc(ngpus, sizeof(network*));
srand(time(0));
int seed = rand();
for(i = 0; i < ngpus; ++i){
srand(seed);
#ifdef GPU
cuda_set_device(gpus[i]);
#endif
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i]->learning_rate *= ngpus;
}
srand(time(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 imgs = net->batch * net->subdivisions * ngpus;
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 *train_list = option_find_str(options, "train", "data/train.list");
list *plist = get_paths(train_list);
char **paths = (char **)list_to_array(plist);
printf("%d\n", plist->size);
int N = plist->size;
load_args args = {0};
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.classes = 80;
args.paths = paths;
args.n = imgs;
args.m = N;
args.type = SEGMENTATION_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){
double time = what_time_is_it_now();
pthread_join(load_thread, 0);
train = buffer;
load_thread = load_data(args);
printf("Loaded: %lf seconds\n", what_time_is_it_now()-time);
time = what_time_is_it_now();
float loss = 0;
#ifdef GPU
if(ngpus == 1){
loss = train_network(net, train);
} else {
loss = train_networks(nets, ngpus, train, 4);
}
#else
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 mask = mask_to_rgb(tr);
image prmask = mask_to_rgb(pred);
show_image(im, "input", 1);
show_image(prmask, "pred", 1);
show_image(mask, "truth", 100);
free_image(mask);
free_image(prmask);
}
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);
free_data(train);
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);
for(i = 0; i < ngpus; ++i){
free_network(nets[i]);
}
free(nets);
free_ptrs((void**)paths, plist->size);
free_list(plist);
free(base);
}
