float cuda_mag_array(float *x_gpu, size_t n) {
float *temp = calloc(n, sizeof(float));
cuda_pull_array(x_gpu, temp, n);
float m = mag_array(temp, n);
free(temp);
return m;
}
image random_unit_vector_image(int w, int h, int c)
{
image im = make_image(w, h, c);
int i;
for(i = 0; i < im.w*im.h*im.c; ++i){
im.data[i] = rand_normal();
}
float mag = mag_array(im.data, im.w*im.h*im.c);
scale_array(im.data, im.w*im.h*im.c, 1./mag);
return im;
}
void slerp(float *start, float *end, float s, int n, float *out)
{
float omega = acos(dot_cpu(n, start, 1, end, 1));
float so = sin(omega);
fill_cpu(n, 0, out, 1);
axpy_cpu(n, sin((1-s)*omega)/so, start, 1, out, 1);
axpy_cpu(n, sin(s*omega)/so, end, 1, out, 1);
float mag = mag_array(out, n);
scale_array(out, n, 1./mag);
}
void test_resize(char *filename)
{
image im = load_image(filename, 0,0, 3);
float mag = mag_array(im.data, im.w*im.h*im.c);
printf("L2 Norm: %f\n", mag);
image gray = grayscale_image(im);
image sat2 = copy_image(im);
saturate_image(sat2, 2);
image sat5 = copy_image(im);
saturate_image(sat5, .5);
image exp2 = copy_image(im);
exposure_image(exp2, 2);
image exp5 = copy_image(im);
exposure_image(exp5, .5);
#ifdef GPU
image r = resize_image(im, im.w, im.h);
image black = make_image(im.w*2 + 3, im.h*2 + 3, 9);
image black2 = make_image(im.w, im.h, 3);
float *r_gpu = cuda_make_array(r.data, r.w*r.h*r.c);
float *black_gpu = cuda_make_array(black.data, black.w*black.h*black.c);
float *black2_gpu = cuda_make_array(black2.data, black2.w*black2.h*black2.c);
shortcut_gpu(3, r.w, r.h, 1, r_gpu, black.w, black.h, 3, black_gpu);
//flip_image(r);
//shortcut_gpu(3, r.w, r.h, 1, r.data, black.w, black.h, 3, black.data);
shortcut_gpu(3, black.w, black.h, 3, black_gpu, black2.w, black2.h, 1, black2_gpu);
cuda_pull_array(black_gpu, black.data, black.w*black.h*black.c);
cuda_pull_array(black2_gpu, black2.data, black2.w*black2.h*black2.c);
show_image_layers(black, "Black");
show_image(black2, "Recreate");
#endif
show_image(im, "Original");
show_image(gray, "Gray");
show_image(sat2, "Saturation-2");
show_image(sat5, "Saturation-.5");
show_image(exp2, "Exposure-2");
show_image(exp5, "Exposure-.5");
#ifdef OPENCV
cvWaitKey(0);
#endif
}
void forward_region_layer(const region_layer l, network_state state)
{
int i,j,b,t,n;
int size = l.coords + l.classes + 1;
memcpy(l.output, state.input, l.outputs*l.batch*sizeof(float));
#ifndef GPU
flatten(l.output, l.w*l.h, size*l.n, l.batch, 1);
#endif
for (b = 0; b < l.batch; ++b){
for(i = 0; i < l.h*l.w*l.n; ++i){
int index = size*i + b*l.outputs;
l.output[index + 4] = logistic_activate(l.output[index + 4]);
}
}
#ifndef GPU
if (l.softmax_tree){
for (b = 0; b < l.batch; ++b){
for(i = 0; i < l.h*l.w*l.n; ++i){
int index = size*i + b*l.outputs;
softmax_tree(l.output + index + 5, 1, 0, 1, l.softmax_tree, l.output + index + 5);
}
}
} else if (l.softmax){
for (b = 0; b < l.batch; ++b){
for(i = 0; i < l.h*l.w*l.n; ++i){
int index = size*i + b*l.outputs;
softmax(l.output + index + 5, l.classes, 1, l.output + index + 5, 1);
}
}
}
#endif
if(!state.train) return;
memset(l.delta, 0, l.outputs * l.batch * sizeof(float));
float avg_iou = 0;
float recall = 0;
float avg_cat = 0;
float avg_obj = 0;
float avg_anyobj = 0;
int count = 0;
int class_count = 0;
*(l.cost) = 0;
for (b = 0; b < l.batch; ++b) {
if(l.softmax_tree){
int onlyclass_id = 0;
for(t = 0; t < l.max_boxes; ++t){
box truth = float_to_box(state.truth + t*5 + b*l.truths);
if(!truth.x) break; // continue;
int class_id = state.truth[t*5 + b*l.truths + 4];
float maxp = 0;
int maxi = 0;
if(truth.x > 100000 && truth.y > 100000){
for(n = 0; n < l.n*l.w*l.h; ++n){
int index = size*n + b*l.outputs + 5;
float scale = l.output[index-1];
float p = scale*get_hierarchy_probability(l.output + index, l.softmax_tree, class_id);
if(p > maxp){
maxp = p;
maxi = n;
}
}
int index = size*maxi + b*l.outputs + 5;
delta_region_class(l.output, l.delta, index, class_id, l.classes, l.softmax_tree, l.class_scale, &avg_cat, l.focal_loss);
++class_count;
onlyclass_id = 1;
break;
}
}
if(onlyclass_id) continue;
}
for (j = 0; j < l.h; ++j) {
for (i = 0; i < l.w; ++i) {
for (n = 0; n < l.n; ++n) {
int index = size*(j*l.w*l.n + i*l.n + n) + b*l.outputs;
box pred = get_region_box(l.output, l.biases, n, index, i, j, l.w, l.h);
float best_iou = 0;
int best_class_id = -1;
for(t = 0; t < l.max_boxes; ++t){
box truth = float_to_box(state.truth + t*5 + b*l.truths);
int class_id = state.truth[t * 5 + b*l.truths + 4];
if (class_id >= l.classes) continue; // if label contains class_id more than number of classes in the cfg-file
if(!truth.x) break; // continue;
float iou = box_iou(pred, truth);
if (iou > best_iou) {
best_class_id = state.truth[t*5 + b*l.truths + 4];
best_iou = iou;
}
}
avg_anyobj += l.output[index + 4];
l.delta[index + 4] = l.noobject_scale * ((0 - l.output[index + 4]) * logistic_gradient(l.output[index + 4]));
if(l.classfix == -1) l.delta[index + 4] = l.noobject_scale * ((best_iou - l.output[index + 4]) * logistic_gradient(l.output[index + 4]));
else{
if (best_iou > l.thresh) {
l.delta[index + 4] = 0;
if(l.classfix > 0){
delta_region_class(l.output, l.delta, index + 5, best_class_id, l.classes, l.softmax_tree, l.class_scale*(l.classfix == 2 ? l.output[index + 4] : 1), &avg_cat, l.focal_loss);
++class_count;
}
}
}
if(*(state.net.seen) < 12800){
box truth = {0};
truth.x = (i + .5)/l.w;
truth.y = (j + .5)/l.h;
truth.w = l.biases[2*n];
truth.h = l.biases[2*n+1];
if(DOABS){
truth.w = l.biases[2*n]/l.w;
truth.h = l.biases[2*n+1]/l.h;
}
delta_region_box(truth, l.output, l.biases, n, index, i, j, l.w, l.h, l.delta, .01);
}
}
}
}
for(t = 0; t < l.max_boxes; ++t){
box truth = float_to_box(state.truth + t*5 + b*l.truths);
int class_id = state.truth[t * 5 + b*l.truths + 4];
if (class_id >= l.classes) {
printf(" Warning: in txt-labels class_id=%d >= classes=%d in cfg-file. In txt-labels class_id should be [from 0 to %d] \n", class_id, l.classes, l.classes-1);
getchar();
continue; // if label contains class_id more than number of classes in the cfg-file
}
if(!truth.x) break; // continue;
float best_iou = 0;
int best_index = 0;
int best_n = 0;
i = (truth.x * l.w);
j = (truth.y * l.h);
//printf("%d %f %d %f\n", i, truth.x*l.w, j, truth.y*l.h);
box truth_shift = truth;
truth_shift.x = 0;
truth_shift.y = 0;
//printf("index %d %d\n",i, j);
for(n = 0; n < l.n; ++n){
int index = size*(j*l.w*l.n + i*l.n + n) + b*l.outputs;
box pred = get_region_box(l.output, l.biases, n, index, i, j, l.w, l.h);
if(l.bias_match){
pred.w = l.biases[2*n];
pred.h = l.biases[2*n+1];
if(DOABS){
pred.w = l.biases[2*n]/l.w;
pred.h = l.biases[2*n+1]/l.h;
}
}
//printf("pred: (%f, %f) %f x %f\n", pred.x, pred.y, pred.w, pred.h);
pred.x = 0;
pred.y = 0;
float iou = box_iou(pred, truth_shift);
if (iou > best_iou){
best_index = index;
best_iou = iou;
best_n = n;
}
}
//printf("%d %f (%f, %f) %f x %f\n", best_n, best_iou, truth.x, truth.y, truth.w, truth.h);
float iou = delta_region_box(truth, l.output, l.biases, best_n, best_index, i, j, l.w, l.h, l.delta, l.coord_scale);
if(iou > .5) recall += 1;
avg_iou += iou;
//l.delta[best_index + 4] = iou - l.output[best_index + 4];
avg_obj += l.output[best_index + 4];
l.delta[best_index + 4] = l.object_scale * (1 - l.output[best_index + 4]) * logistic_gradient(l.output[best_index + 4]);
if (l.rescore) {
l.delta[best_index + 4] = l.object_scale * (iou - l.output[best_index + 4]) * logistic_gradient(l.output[best_index + 4]);
}
if (l.map) class_id = l.map[class_id];
delta_region_class(l.output, l.delta, best_index + 5, class_id, l.classes, l.softmax_tree, l.class_scale, &avg_cat, l.focal_loss);
++count;
++class_count;
}
}
//printf("\n");
#ifndef GPU
flatten(l.delta, l.w*l.h, size*l.n, l.batch, 0);
#endif
*(l.cost) = pow(mag_array(l.delta, l.outputs * l.batch), 2);
printf("Region Avg IOU: %f, Class: %f, Obj: %f, No Obj: %f, Avg Recall: %f, count: %d\n", avg_iou/count, avg_cat/class_count, avg_obj/count, avg_anyobj/(l.w*l.h*l.n*l.batch), recall/count, count);
}
void forward_detection_layer(const detection_layer l, network_state state)
{
int locations = l.side*l.side;
int i,j;
memcpy(l.output, state.input, l.outputs*l.batch*sizeof(float));
int b;
if (l.softmax){
for(b = 0; b < l.batch; ++b){
int index = b*l.inputs;
for (i = 0; i < locations; ++i) {
int offset = i*l.classes;
softmax_array(l.output + index + offset, l.classes, 1,
l.output + index + offset);
}
}
}
if(state.train){
float avg_iou = 0;
float avg_cat = 0;
float avg_allcat = 0;
float avg_obj = 0;
float avg_anyobj = 0;
int count = 0;
*(l.cost) = 0;
int size = l.inputs * l.batch;
memset(l.delta, 0, size * sizeof(float));
for (b = 0; b < l.batch; ++b){
int index = b*l.inputs;
for (i = 0; i < locations; ++i) {
int truth_index = (b*locations + i)*(1+l.coords+l.classes);
int is_obj = state.truth[truth_index];
for (j = 0; j < l.n; ++j) {
int p_index = index + locations*l.classes + i*l.n + j;
l.delta[p_index] = l.noobject_scale*(0 - l.output[p_index]);
*(l.cost) += l.noobject_scale*pow(l.output[p_index], 2);
avg_anyobj += l.output[p_index];
}
int best_index = -1;
float best_iou = 0;
float best_rmse = 20;
if (!is_obj){
continue;
}
int class_index = index + i*l.classes;
for(j = 0; j < l.classes; ++j) {
l.delta[class_index+j] = l.class_scale * (state.truth[truth_index+1+j] - l.output[class_index+j]);
*(l.cost) += l.class_scale * pow(state.truth[truth_index+1+j] - l.output[class_index+j], 2);
if(state.truth[truth_index + 1 + j]) avg_cat += l.output[class_index+j];
avg_allcat += l.output[class_index+j];
}
box truth = float_to_box(state.truth + truth_index + 1 + l.classes);
truth.x /= l.side;
truth.y /= l.side;
for(j = 0; j < l.n; ++j){
int box_index = index + locations*(l.classes + l.n) + (i*l.n + j) * l.coords;
box out = float_to_box(l.output + box_index);
out.x /= l.side;
out.y /= l.side;
if (l.sqrt){
out.w = out.w*out.w;
out.h = out.h*out.h;
}
float iou = box_iou(out, truth);
//iou = 0;
float rmse = box_rmse(out, truth);
if(best_iou > 0 || iou > 0){
if(iou > best_iou){
best_iou = iou;
best_index = j;
}
}else{
if(rmse < best_rmse){
best_rmse = rmse;
best_index = j;
}
}
}
if(l.forced){
if(truth.w*truth.h < .1){
best_index = 1;
}else{
best_index = 0;
}
}
if(l.random && *(state.net.seen) < 64000){
best_index = rand()%l.n;
}
int box_index = index + locations*(l.classes + l.n) + (i*l.n + best_index) * l.coords;
int tbox_index = truth_index + 1 + l.classes;
box out = float_to_box(l.output + box_index);
out.x /= l.side;
out.y /= l.side;
if (l.sqrt) {
out.w = out.w*out.w;
out.h = out.h*out.h;
}
float iou = box_iou(out, truth);
//printf("%d,", best_index);
int p_index = index + locations*l.classes + i*l.n + best_index;
*(l.cost) -= l.noobject_scale * pow(l.output[p_index], 2);
*(l.cost) += l.object_scale * pow(1-l.output[p_index], 2);
avg_obj += l.output[p_index];
l.delta[p_index] = l.object_scale * (1.-l.output[p_index]);
if(l.rescore){
l.delta[p_index] = l.object_scale * (iou - l.output[p_index]);
}
l.delta[box_index+0] = l.coord_scale*(state.truth[tbox_index + 0] - l.output[box_index + 0]);
l.delta[box_index+1] = l.coord_scale*(state.truth[tbox_index + 1] - l.output[box_index + 1]);
l.delta[box_index+2] = l.coord_scale*(state.truth[tbox_index + 2] - l.output[box_index + 2]);
l.delta[box_index+3] = l.coord_scale*(state.truth[tbox_index + 3] - l.output[box_index + 3]);
if(l.sqrt){
l.delta[box_index+2] = l.coord_scale*(sqrt(state.truth[tbox_index + 2]) - l.output[box_index + 2]);
l.delta[box_index+3] = l.coord_scale*(sqrt(state.truth[tbox_index + 3]) - l.output[box_index + 3]);
}
*(l.cost) += pow(1-iou, 2);
avg_iou += iou;
++count;
}
}
if(0){
float *costs = calloc(l.batch*locations*l.n, sizeof(float));
for (b = 0; b < l.batch; ++b) {
int index = b*l.inputs;
for (i = 0; i < locations; ++i) {
for (j = 0; j < l.n; ++j) {
int p_index = index + locations*l.classes + i*l.n + j;
costs[b*locations*l.n + i*l.n + j] = l.delta[p_index]*l.delta[p_index];
}
}
}
int indexes[100];
top_k(costs, l.batch*locations*l.n, 100, indexes);
float cutoff = costs[indexes[99]];
for (b = 0; b < l.batch; ++b) {
int index = b*l.inputs;
for (i = 0; i < locations; ++i) {
for (j = 0; j < l.n; ++j) {
int p_index = index + locations*l.classes + i*l.n + j;
if (l.delta[p_index]*l.delta[p_index] < cutoff) l.delta[p_index] = 0;
}
}
}
free(costs);
}
*(l.cost) = pow(mag_array(l.delta, l.outputs * l.batch), 2);
if ( l.b_debug )
{
printf("Detection Avg IOU: %f, Pos Cat: %f, All Cat: %f, Pos Obj: %f, Any Obj: %f, count: %d\n", avg_iou/count, avg_cat/count, avg_allcat/(count*l.classes), avg_obj/count, avg_anyobj/(l.batch*locations*l.n), count);
}
}
}
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 forward_iseg_layer(const layer l, network net) {
double time = what_time_is_it_now();
int i, b, j, k;
int ids = l.extra;
memcpy(l.output, net.input, l.outputs * l.batch * sizeof(real_t));
memset(l.delta, 0, l.outputs * l.batch * sizeof(real_t));
#ifndef GPU
for (b = 0; b < l.batch; ++b) {
int index = b * l.outputs;
activate_array(l.output + index, l.classes * l.w * l.h, LOGISTIC);
}
#endif
for (b = 0; b < l.batch; ++b) {
// a priori, each pixel has no class
for (i = 0; i < l.classes; ++i) {
for (k = 0; k < l.w * l.h; ++k) {
int index = b * l.outputs + i * l.w * l.h + k;
l.delta[index] = 0 - l.output[index];
}
}
// a priori, embedding should be small magnitude
for (i = 0; i < ids; ++i) {
for (k = 0; k < l.w * l.h; ++k) {
int index = b * l.outputs + (i + l.classes) * l.w * l.h + k;
l.delta[index] = .1 * (0 - l.output[index]);
}
}
memset(l.counts, 0, 90 * sizeof(int));
for (i = 0; i < 90; ++i) {
fill_cpu(ids, 0, l.sums[i], 1);
int c = net.truth[b * l.truths + i * (l.w * l.h + 1)];
if (c < 0)
break;
// add up metric embeddings for each instance
for (k = 0; k < l.w * l.h; ++k) {
int index = b * l.outputs + c * l.w * l.h + k;
real_t v = net.truth[b * l.truths + i * (l.w * l.h + 1) + 1 + k];
if (v) {
l.delta[index] = v - l.output[index];
axpy_cpu(ids, 1,
l.output + b * l.outputs + l.classes * l.w * l.h
+ k, l.w * l.h, l.sums[i], 1);
++l.counts[i];
}
}
}
real_t *mse = calloc(90, sizeof(real_t));
for (i = 0; i < 90; ++i) {
int c = net.truth[b * l.truths + i * (l.w * l.h + 1)];
if (c < 0)
break;
for (k = 0; k < l.w * l.h; ++k) {
real_t v = net.truth[b * l.truths + i * (l.w * l.h + 1) + 1 + k];
if (v) {
int z;
real_t sum = 0;
for (z = 0; z < ids; ++z) {
int index = b * l.outputs + (l.classes + z) * l.w * l.h
+ k;
sum += pow(l.sums[i][z] / l.counts[i] - l.output[index],
2);
}
mse[i] += sum;
}
}
mse[i] /= l.counts[i];
}
// Calculate average embedding
for (i = 0; i < 90; ++i) {
if (!l.counts[i])
continue;
scal_cpu(ids, 1.f / l.counts[i], l.sums[i], 1);
if (b == 0 && net.gpu_index == 0) {
printf("%4d, %6.3f, ", l.counts[i], mse[i]);
for (j = 0; j < ids; ++j) {
printf("%6.3f,", l.sums[i][j]);
}
printf("\n");
}
}
free(mse);
// Calculate embedding loss
for (i = 0; i < 90; ++i) {
if (!l.counts[i])
continue;
for (k = 0; k < l.w * l.h; ++k) {
real_t v = net.truth[b * l.truths + i * (l.w * l.h + 1) + 1 + k];
if (v) {
for (j = 0; j < 90; ++j) {
if (!l.counts[j])
continue;
int z;
for (z = 0; z < ids; ++z) {
int index = b * l.outputs
+ (l.classes + z) * l.w * l.h + k;
real_t diff = l.sums[j][z] - l.output[index];
if (j == i)
l.delta[index] += diff < 0 ? -.1 : .1;
else
l.delta[index] += -(diff < 0 ? -.1 : .1);
}
}
}
}
}
for (i = 0; i < ids; ++i) {
for (k = 0; k < l.w * l.h; ++k) {
int index = b * l.outputs + (i + l.classes) * l.w * l.h + k;
l.delta[index] *= .01;
}
}
}
*(l.cost) = pow(mag_array(l.delta, l.outputs * l.batch), 2);
printf("took %lf sec\n", what_time_is_it_now() - time);
}