caf::behavior init_buffer() {
// Initial send tasks
auto pending_chunks = std::make_shared<int>(workers_.size());
auto pending_jobs = std::make_shared<int>(buffer_min_size_);
send_job();
return {
[=](const std::vector<uint16_t>& data, uint32_t id) {
concat_data(data, id);
if (! --*pending_chunks) {
if (--*pending_jobs) {
send_job();
*pending_chunks = workers_.size();
} else {
become(main_phase());
}
}
},
[=](resize_atom, uint32_t w, uint32_t h) {
resize(w,h);
},
[=](limit_atom, normal_atom, uint32_t workers) {
become(make_behavior());
send(this, calc_weights_atom::value, size_t{workers});
},
caf::others() >> [=] {
std::cout << to_string(current_message()) << std::endl;
}
};
}
caf::behavior main_phase() {
send(this, tick_atom::value);
return {
[=](const std::vector<uint16_t>& data, uint32_t id) {
concat_data(data, id);
},
[=](tick_atom) {
delayed_send(this, tick_rate_, tick_atom::value);
if (cache_.empty()) {
std::cout << "[WARNING] Cache empty..." << std::endl;
return;
}
send(sink_, image_width_, cache_.front());
cache_.pop();
send_job();
},
[=](resize_atom, uint32_t w, uint32_t h) {
resize(w,h);
},
[=](limit_atom, normal_atom, uint32_t workers) {
become(make_behavior());
send(this, calc_weights_atom::value, size_t{workers});
},
caf::others() >> [=] {
std::cout << to_string(current_message()) << std::endl;
}
};
}
hash_digest build_merkle_tree(hash_list& merkle)
{
// Stop if hash list is empty.
if (merkle.empty())
return null_hash;
else if (merkle.size() == 1)
return merkle[0];
// While there is more than 1 hash in the list, keep looping...
while (merkle.size() > 1)
{
// If number of hashes is odd, duplicate last hash in the list.
if (merkle.size() % 2 != 0)
merkle.push_back(merkle.back());
// List size is now even.
BITCOIN_ASSERT(merkle.size() % 2 == 0);
// New hash list.
hash_list new_merkle;
// Loop through hashes 2 at a time.
for (auto it = merkle.begin(); it != merkle.end(); it += 2)
{
// Join both current hashes together (concatenate).
data_chunk concat_data(hash_size * 2);
auto concat = make_serializer(concat_data.begin());
concat.write_hash(*it);
concat.write_hash(*(it + 1));
BITCOIN_ASSERT(concat.iterator() == concat_data.end());
// Hash both of the hashes.
hash_digest new_root = bitcoin_hash(concat_data);
// Add this to the new list.
new_merkle.push_back(new_root);
}
// This is the new list.
merkle = new_merkle;
}
// Finally we end up with a single item.
return merkle[0];
}
void train_colorizer(char *cfg, char *weight, char *acfg, char *aweight, int clear, int display)
{
#ifdef GPU
//char *train_images = "/home/kunle12/data/coco/train1.txt";
//char *train_images = "/home/kunle12/data/coco/trainvalno5k.txt";
char *train_images = "/home/kunle12/data/imagenet/imagenet1k.train.list";
char *backup_directory = "/home/kunle12/backup/";
srand(time(0));
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);
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];
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;
data train, buffer;
list *plist = get_paths(train_images);
//int N = plist->size;
char **paths = (char **)list_to_array(plist);
load_args args= get_base_args(net);
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.d = &buffer;
args.type = CLASSIFICATION_DATA;
args.classes = 1;
char *ls[2] = {"imagenet"};
args.labels = ls;
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
int x_size = net->inputs*net->batch;
//int y_size = x_size;
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;
float *imerror = cuda_make_array(0, imlayer.outputs*imlayer.batch);
float aloss_avg = -1;
float gloss_avg = -1;
//data generated = copy_data(train);
while (get_current_batch(net) < net->max_batches) {
i += 1;
time=clock();
pthread_join(load_thread, 0);
train = buffer;
load_thread = load_data_in_thread(args);
printf("Loaded: %lf seconds\n", sec(clock()-time));
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]);
grayscale_image_3c(gim);
train.y.vals[j][0] = .95;
gray.y.vals[j][0] = .05;
}
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);
/*
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;
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;
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, 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));
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);
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);
}
}
harmless_update_network_gpu(anet);
data merge = concat_data(train, gray);
//randomize_data(merge);
float aloss = train_network(anet, merge);
update_network_gpu(net);
#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]);
show_image(im, "gen", 1);
show_image(im2, "train", 1);
}
#endif
free_data(merge);
free_data(train);
free_data(gray);
if (aloss_avg < 0) aloss_avg = aloss;
aloss_avg = aloss_avg*.9 + aloss*.1;
gloss_avg = gloss_avg*.9 + gloss*.1;
printf("%d: gen: %f, adv: %f | gen_avg: %f, adv_avg: %f, %f rate, %lf seconds, %d images\n", i, gloss, aloss, gloss_avg, aloss_avg, get_current_rate(net), sec(clock()-time), i*imgs);
if(i%1000==0){
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
save_weights(net, buff);
sprintf(buff, "%s/%s_%d.weights", backup_directory, abase, i);
save_weights(anet, buff);
}
if(i%100==0){
char buff[256];
sprintf(buff, "%s/%s.backup", backup_directory, base);
save_weights(net, buff);
sprintf(buff, "%s/%s.backup", backup_directory, abase);
save_weights(anet, buff);
}
}
char buff[256];
sprintf(buff, "%s/%s_final.weights", backup_directory, base);
save_weights(net, buff);
#endif
}
void train_dcgan(char *cfg, char *weight, char *acfg, char *aweight, int clear, int display, char *train_images, int maxbatch)
{
#ifdef GPU
char *backup_directory = "/home/kunle12/backup/";
srand(time(0));
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;
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];
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;
data train, buffer;
list *plist = get_paths(train_images);
//int N = plist->size;
char **paths = (char **)list_to_array(plist);
load_args args= get_base_args(anet);
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.d = &buffer;
args.type = CLASSIFICATION_DATA;
args.threads=16;
args.classes = 1;
char *ls[2] = {"imagenet", "zzzzzzzz"};
args.labels = ls;
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
gnet->train = 1;
anet->train = 1;
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;
float aloss_avg = -1;
//data generated = copy_data(train);
if (maxbatch == 0) maxbatch = gnet->max_batches;
while (get_current_batch(gnet) < maxbatch) {
i += 1;
time=clock();
pthread_join(load_thread, 0);
train = buffer;
//translate_data_rows(train, -.5);
//scale_data_rows(train, 2);
load_thread = load_data_in_thread(args);
printf("Loaded: %lf seconds\n", sec(clock()-time));
data gen = copy_data(train);
for (j = 0; j < imgs; ++j) {
train.y.vals[j][0] = 1;
gen.y.vals[j][0] = 0;
}
time=clock();
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();
}
for(z = 0; z < gnet->batch; ++z){
float mag = mag_array(gnet->input + z*gnet->inputs, gnet->inputs);
scale_array(gnet->input + z*gnet->inputs, gnet->inputs, 1./mag);
}
/*
for(z = 0; z < 100; ++z){
printf("%f, ", gnet->input[z]);
}
printf("\n");
printf("input: %f %f\n", mean_array(gnet->input, x_size), variance_array(gnet->input, x_size));
*/
//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(gnet);
fill_gpu(imlayer.outputs*imlayer.batch, 0, imerror, 1);
fill_cpu(anet->truths*anet->batch, 1, anet->truth, 1);
copy_cpu(anet->inputs*anet->batch, imlayer.output, 1, anet->input, 1);
anet->delta_gpu = imerror;
forward_network(anet);
backward_network(anet);
//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, 0, 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));
axpy_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1, gnet->layers[gnet->n-1].delta_gpu, 1);
backward_network(gnet);
/*
for(k = 0; k < gnet->n; ++k){
layer l = gnet->layers[k];
cuda_pull_array(l.output_gpu, l.output, l.outputs*l.batch);
printf("%d: %f %f\n", k, mean_array(l.output, l.outputs*l.batch), variance_array(l.output, l.outputs*l.batch));
}
*/
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);
data merge = concat_data(train, gen);
//randomize_data(merge);
float aloss = train_network(anet, merge);
//translate_image(im, 1);
//scale_image(im, .5);
//translate_image(im2, 1);
//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]);
show_image(im, "gen", 1);
show_image(im2, "train", 1);
save_image(im, "gen");
save_image(im2, "train");
}
#endif
/*
if(aloss < .1){
anet->learning_rate = 0;
} else if (aloss > .3){
anet->learning_rate = orig_rate;
}
*/
update_network_gpu(gnet);
free_data(merge);
free_data(train);
free_data(gen);
if (aloss_avg < 0) aloss_avg = aloss;
aloss_avg = aloss_avg*.9 + aloss*.1;
printf("%d: adv: %f | adv_avg: %f, %f rate, %lf seconds, %d images\n", i, aloss, aloss_avg, get_current_rate(gnet), sec(clock()-time), i*imgs);
if(i%10000==0){
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
save_weights(gnet, buff);
sprintf(buff, "%s/%s_%d.weights", backup_directory, abase, i);
save_weights(anet, buff);
}
if(i%1000==0){
char buff[256];
sprintf(buff, "%s/%s.backup", backup_directory, base);
save_weights(gnet, buff);
sprintf(buff, "%s/%s.backup", backup_directory, abase);
save_weights(anet, buff);
}
}
char buff[256];
sprintf(buff, "%s/%s_final.weights", backup_directory, base);
save_weights(gnet, buff);
#endif
free_network(gnet);
free_network(anet);
}
void train_prog(char *cfg, char *weight, char *acfg, char *aweight, int clear, int display, char *train_images, int maxbatch)
{
#ifdef GPU
char *backup_directory = "/home/kunle12/backup/";
srand(time(0));
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);
int i, j, k;
layer imlayer = gnet->layers[gnet->n-1];
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;
list *plist = get_paths(train_images);
char **paths = (char **)list_to_array(plist);
load_args args= get_base_args(anet);
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.d = &buffer;
args.type = CLASSIFICATION_DATA;
args.threads=16;
args.classes = 1;
char *ls[2] = {"imagenet", "zzzzzzzz"};
args.labels = ls;
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
gnet->train = 1;
anet->train = 1;
int x_size = gnet->inputs*gnet->batch;
int y_size = gnet->truths*gnet->batch;
float *imerror = cuda_make_array(0, y_size);
float aloss_avg = -1;
if (maxbatch == 0) maxbatch = gnet->max_batches;
while (get_current_batch(gnet) < maxbatch) {
{
int cb = get_current_batch(gnet);
float alpha = (float) cb / (maxbatch/2);
if(alpha > 1) alpha = 1;
float beta = 1 - alpha;
printf("%f %f\n", alpha, beta);
set_network_alpha_beta(gnet, alpha, beta);
set_network_alpha_beta(anet, beta, alpha);
}
i += 1;
time=clock();
pthread_join(load_thread, 0);
train = buffer;
load_thread = load_data_in_thread(args);
printf("Loaded: %lf seconds\n", sec(clock()-time));
data gen = copy_data(train);
for (j = 0; j < imgs; ++j) {
train.y.vals[j][0] = 1;
gen.y.vals[j][0] = 0;
}
time=clock();
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();
}
/*
for(z = 0; z < gnet->batch; ++z){
float mag = mag_array(gnet->input + z*gnet->inputs, gnet->inputs);
scale_array(gnet->input + z*gnet->inputs, gnet->inputs, 1./mag);
}
*/
*gnet->seen += gnet->batch;
forward_network(gnet);
fill_gpu(imlayer.outputs*imlayer.batch, 0, imerror, 1);
fill_cpu(anet->truths*anet->batch, 1, anet->truth, 1);
copy_cpu(anet->inputs*anet->batch, imlayer.output, 1, anet->input, 1);
anet->delta_gpu = imerror;
forward_network(anet);
backward_network(anet);
//float genaloss = *anet->cost / anet->batch;
scal_gpu(imlayer.outputs*imlayer.batch, 1, imerror, 1);
scal_gpu(imlayer.outputs*imlayer.batch, 0, 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(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);
}
}
harmless_update_network_gpu(anet);
data merge = concat_data(train, gen);
float aloss = train_network(anet, merge);
#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]);
show_image(im, "gen", 1);
show_image(im2, "train", 1);
save_image(im, "gen");
save_image(im2, "train");
}
#endif
update_network_gpu(gnet);
free_data(merge);
free_data(train);
free_data(gen);
if (aloss_avg < 0) aloss_avg = aloss;
aloss_avg = aloss_avg*.9 + aloss*.1;
printf("%d: adv: %f | adv_avg: %f, %f rate, %lf seconds, %d images\n", i, aloss, aloss_avg, get_current_rate(gnet), sec(clock()-time), i*imgs);
if(i%10000==0){
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
save_weights(gnet, buff);
sprintf(buff, "%s/%s_%d.weights", backup_directory, abase, i);
save_weights(anet, buff);
}
if(i%1000==0){
char buff[256];
sprintf(buff, "%s/%s.backup", backup_directory, base);
save_weights(gnet, buff);
sprintf(buff, "%s/%s.backup", backup_directory, abase);
save_weights(anet, buff);
}
}
char buff[256];
sprintf(buff, "%s/%s_final.weights", backup_directory, base);
save_weights(gnet, buff);
#endif
free_network( gnet );
free_network( anet );
}
void train_lsd2(char *cfgfile, char *weightfile, char *acfgfile, char *aweightfile, int clear)
{
#ifdef GPU
char *train_images = "/home/pjreddie/data/coco/trainvalno5k.txt";
char *backup_directory = "/home/pjreddie/backup/";
srand(time(0));
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;
char *abase = basecfg(acfgfile);
network anet = parse_network_cfg(acfgfile);
if(aweightfile){
load_weights(&anet, aweightfile);
}
if(clear) *anet.seen = 0;
int i, j, k;
layer imlayer = {};
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;
data train, buffer;
list *plist = get_paths(train_images);
//int N = plist->size;
char **paths = (char **)list_to_array(plist);
load_args args = {};
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.type = CLASSIFICATION_DATA;
args.classes = 1;
char *ls[1] = {"coco"};
args.labels = ls;
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
network_state gstate = {};
gstate.index = 0;
gstate.net = net;
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;
gstate.train = 1;
float *X = (float*)calloc(x_size, sizeof(float));
float *y = (float*)calloc(y_size, sizeof(float));
network_state astate = {};
astate.index = 0;
astate.net = anet;
int ay_size = get_network_output_size(anet)*anet.batch;
astate.input = 0;
astate.truth = 0;
astate.delta = 0;
astate.train = 1;
float *imerror = cuda_make_array(0, imlayer.outputs);
float *ones_gpu = cuda_make_array(0, ay_size);
fill_ongpu(ay_size, 1, ones_gpu, 1);
float aloss_avg = -1;
float gloss_avg = -1;
//data generated = copy_data(train);
while (get_current_batch(net) < net.max_batches) {
i += 1;
time=clock();
pthread_join(load_thread, 0);
train = buffer;
load_thread = load_data_in_thread(args);
printf("Loaded: %lf seconds\n", sec(clock()-time));
data generated = copy_data(train);
time=clock();
float gloss = 0;
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;
forward_network_gpu(net, gstate);
fill_ongpu(imlayer.outputs, 0, imerror, 1);
astate.input = imlayer.output_gpu;
astate.delta = imerror;
astate.truth = ones_gpu;
forward_network_gpu(anet, astate);
backward_network_gpu(anet, astate);
scal_ongpu(imlayer.outputs, 1, imerror, 1);
axpy_ongpu(imlayer.outputs, 1, imerror, 1, imlayer.delta_gpu, 1);
backward_network_gpu(net, gstate);
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);
cuda_pull_array(imlayer.output_gpu, imlayer.output, x_size);
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;
}
}
harmless_update_network_gpu(anet);
data merge = concat_data(train, generated);
randomize_data(merge);
float aloss = train_network(anet, merge);
update_network_gpu(net);
update_network_gpu(anet);
free_data(merge);
free_data(train);
free_data(generated);
if (aloss_avg < 0) aloss_avg = aloss;
aloss_avg = aloss_avg*.9 + aloss*.1;
gloss_avg = gloss_avg*.9 + gloss*.1;
printf("%d: gen: %f, adv: %f | gen_avg: %f, adv_avg: %f, %f rate, %lf seconds, %d images\n", i, gloss, aloss, gloss_avg, aloss_avg, get_current_rate(net), sec(clock()-time), i*imgs);
if(i%1000==0){
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
save_weights(net, buff);
sprintf(buff, "%s/%s_%d.weights", backup_directory, abase, i);
save_weights(anet, buff);
}
if(i%100==0){
char buff[256];
sprintf(buff, "%s/%s.backup", backup_directory, base);
save_weights(net, buff);
sprintf(buff, "%s/%s.backup", backup_directory, abase);
save_weights(anet, buff);
}
}
char buff[256];
sprintf(buff, "%s/%s_final.weights", backup_directory, base);
save_weights(net, buff);
#endif
}
void train_lsd3(char *fcfg, char *fweight, char *gcfg, char *gweight, char *acfg, char *aweight, int clear)
{
#ifdef GPU
//char *train_images = "/home/pjreddie/data/coco/trainvalno5k.txt";
char *train_images = "/home/pjreddie/data/imagenet/imagenet1k.train.list";
//char *style_images = "/home/pjreddie/data/coco/trainvalno5k.txt";
char *style_images = "/home/pjreddie/zelda.txt";
char *backup_directory = "/home/pjreddie/backup/";
srand(time(0));
network fnet = load_network(fcfg, fweight, clear);
network gnet = load_network(gcfg, gweight, clear);
network anet = load_network(acfg, aweight, clear);
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;
data train, tbuffer;
data style, sbuffer;
list *slist = get_paths(style_images);
char **spaths = (char **)list_to_array(slist);
list *tlist = get_paths(train_images);
char **tpaths = (char **)list_to_array(tlist);
load_args targs= get_base_args(gnet);
targs.paths = tpaths;
targs.n = imgs;
targs.m = tlist->size;
targs.d = &tbuffer;
targs.type = CLASSIFICATION_DATA;
targs.classes = 1;
char *ls[1] = {"zelda"};
targs.labels = ls;
load_args sargs = get_base_args(gnet);
sargs.paths = spaths;
sargs.n = imgs;
sargs.m = slist->size;
sargs.d = &sbuffer;
sargs.type = CLASSIFICATION_DATA;
sargs.classes = 1;
sargs.labels = ls;
pthread_t tload_thread = load_data_in_thread(targs);
pthread_t sload_thread = load_data_in_thread(sargs);
clock_t time;
float aloss_avg = -1;
float floss_avg = -1;
network_state fstate = {};
fstate.index = 0;
fstate.net = fnet;
int x_size = get_network_input_size(fnet)*fnet.batch;
int y_size = get_network_output_size(fnet)*fnet.batch;
fstate.input = cuda_make_array(0, x_size);
fstate.truth = cuda_make_array(0, y_size);
fstate.delta = cuda_make_array(0, x_size);
fstate.train = 1;
float *X = (float*)calloc(x_size, sizeof(float));
float *y = (float*)calloc(y_size, sizeof(float));
float *ones = cuda_make_array(0, anet.batch);
float *zeros = cuda_make_array(0, anet.batch);
fill_ongpu(anet.batch, .99, ones, 1);
fill_ongpu(anet.batch, .01, zeros, 1);
network_state astate = {};
astate.index = 0;
astate.net = anet;
int ax_size = get_network_input_size(anet)*anet.batch;
int ay_size = get_network_output_size(anet)*anet.batch;
astate.input = 0;
astate.truth = ones;
astate.delta = cuda_make_array(0, ax_size);
astate.train = 1;
network_state gstate = {};
gstate.index = 0;
gstate.net = gnet;
int gx_size = get_network_input_size(gnet)*gnet.batch;
int gy_size = get_network_output_size(gnet)*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) {
i += 1;
time=clock();
pthread_join(tload_thread, 0);
pthread_join(sload_thread, 0);
train = tbuffer;
style = sbuffer;
tload_thread = load_data_in_thread(targs);
sload_thread = load_data_in_thread(sargs);
printf("Loaded: %lf seconds\n", sec(clock()-time));
data generated = copy_data(train);
time=clock();
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);
cuda_push_array(fstate.input, X, x_size);
cuda_push_array(gstate.input, X, gx_size);
*gnet.seen += gnet.batch;
forward_network_gpu(fnet, fstate);
float *feats = fnet.layers[fnet.n - 2].output_gpu;
copy_ongpu(y_size, feats, 1, fstate.truth, 1);
forward_network_gpu(gnet, gstate);
float *gen = gnet.layers[gnet.n-1].output_gpu;
copy_ongpu(x_size, gen, 1, fstate.input, 1);
fill_ongpu(x_size, 0, fstate.delta, 1);
forward_network_gpu(fnet, fstate);
backward_network_gpu(fnet, fstate);
//HERE
astate.input = gen;
fill_ongpu(ax_size, 0, astate.delta, 1);
forward_network_gpu(anet, astate);
backward_network_gpu(anet, astate);
float *delta = imlayer.delta_gpu;
fill_ongpu(x_size, 0, delta, 1);
scal_ongpu(x_size, 100, astate.delta, 1);
scal_ongpu(x_size, .00001, fstate.delta, 1);
axpy_ongpu(x_size, 1, fstate.delta, 1, delta, 1);
axpy_ongpu(x_size, 1, astate.delta, 1, delta, 1);
//fill_ongpu(x_size, 0, delta, 1);
//cuda_push_array(delta, X, x_size);
//axpy_ongpu(x_size, -1, imlayer.output_gpu, 1, delta, 1);
//printf("pix error: %f\n", cuda_mag_array(delta, x_size));
printf("fea error: %f\n", cuda_mag_array(fstate.delta, x_size));
printf("adv error: %f\n", cuda_mag_array(astate.delta, x_size));
//axpy_ongpu(x_size, 1, astate.delta, 1, delta, 1);
backward_network_gpu(gnet, gstate);
floss += get_network_cost(fnet) /(fnet.subdivisions*fnet.batch);
cuda_pull_array(imlayer.output_gpu, imlayer.output, x_size);
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] = .01;
}
}
/*
image sim = float_to_image(anet.w, anet.h, anet.c, style.X.vals[j]);
show_image(sim, "style");
cvWaitKey(0);
*/
harmless_update_network_gpu(anet);
data merge = concat_data(style, generated);
randomize_data(merge);
float aloss = train_network(anet, merge);
update_network_gpu(gnet);
free_data(merge);
free_data(train);
free_data(generated);
free_data(style);
if (aloss_avg < 0) aloss_avg = aloss;
if (floss_avg < 0) floss_avg = floss;
aloss_avg = aloss_avg*.9 + aloss*.1;
floss_avg = floss_avg*.9 + floss*.1;
printf("%d: gen: %f, adv: %f | gen_avg: %f, adv_avg: %f, %f rate, %lf seconds, %d images\n", i, floss, aloss, floss_avg, aloss_avg, get_current_rate(gnet), sec(clock()-time), i*imgs);
if(i%1000==0){
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, gbase, i);
save_weights(gnet, buff);
sprintf(buff, "%s/%s_%d.weights", backup_directory, abase, i);
save_weights(anet, buff);
}
if(i%100==0){
char buff[256];
sprintf(buff, "%s/%s.backup", backup_directory, gbase);
save_weights(gnet, buff);
sprintf(buff, "%s/%s.backup", backup_directory, abase);
save_weights(anet, buff);
}
}
#endif
}