github
/
darknet
mirror of https://github.com/AlexeyAB/darknet.git
1
0
Fork 0
Code Issues Releases Wiki Activity

Fixed many warnings

Browse Source
pull/2383/head^2
AlexeyAB 6 years ago
parent 6d6202f666
commit b3254ed523
20 changed files with 73 additions and 75 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 2
      src/captcha.c
  2. 4
      src/cifar.c
  3. 6
      src/classifier.c
  4. 2
      src/compare.c
  5. 25
      src/convolutional_layer.c
  6. 2
      src/darknet.c
  7. 2
      src/darkunistd.h
  8. 8
      src/data.c
  9. 21
      src/detector.c
  10. 2
      src/dice.c
  11. 28
      src/gemm.c
  12. 2
      src/go.c
  13. 2
      src/http_stream.cpp
  14. 3
      src/matrix.c
  15. 6
      src/network.c
  16. 2
      src/rnn.c
  17. 2
      src/tag.c
  18. 17
      src/utils.c
  19. 2
      src/writing.c
  20. 10
      src/yolo_v2_class.cpp

2
src/captcha.c
Unescape View File

@ -85,7 +85,7 @@ void train_captcha(char *cfgfile, char *weightfile)
float loss = train_network(net, train); float loss = train_network(net, train);
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1; 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, %ld images\n", i, loss, avg_loss, sec(clock()-time), *net.seen);
free_data(train); free_data(train);
if(i%100==0){ if(i%100==0){
char buff[256]; char buff[256];

4
src/cifar.c
Unescape View File

@ -33,7 +33,7 @@ void train_cifar(char *cfgfile, char *weightfile)
float loss = train_network_sgd(net, train, 1); float loss = train_network_sgd(net, train, 1);
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.95 + loss*.05; avg_loss = avg_loss*.95 + loss*.05;
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); printf("%d, %.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){ if(*net.seen/N > epoch){
epoch = *net.seen/N; epoch = *net.seen/N;
char buff[256]; char buff[256];
@ -89,7 +89,7 @@ void train_cifar_distill(char *cfgfile, char *weightfile)
float loss = train_network_sgd(net, train, 1); float loss = train_network_sgd(net, train, 1);
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.95 + loss*.05; avg_loss = avg_loss*.95 + loss*.05;
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); printf("%d, %.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){ if(*net.seen/N > epoch){
epoch = *net.seen/N; epoch = *net.seen/N;
char buff[256]; char buff[256];

6
src/classifier.c
Unescape View File

@ -177,7 +177,7 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
draw_precision = 1; draw_precision = 1;
} }
printf("%d, %.3f: %f, %f avg, %f rate, %lf seconds, %d images\n", get_current_batch(net), (float)(*net.seen)/ train_images_num, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen); printf("%d, %.3f: %f, %f avg, %f rate, %lf seconds, %ld images\n", get_current_batch(net), (float)(*net.seen)/ train_images_num, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
#ifdef OPENCV #ifdef OPENCV
draw_train_loss(img, img_size, avg_loss, max_img_loss, i, net.max_batches, topk, draw_precision, "top5", dont_show, mjpeg_port); draw_train_loss(img, img_size, avg_loss, max_img_loss, i, net.max_batches, topk, draw_precision, "top5", dont_show, mjpeg_port);
#endif // OPENCV #endif // OPENCV
@ -198,7 +198,7 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int *gpus,
if (ngpus != 1) sync_nets(nets, ngpus, 0); if (ngpus != 1) sync_nets(nets, ngpus, 0);
#endif #endif
char buff[256]; char buff[256];
sprintf(buff, "%s/%s_last.weights", backup_directory, base, i); sprintf(buff, "%s/%s_last.weights", backup_directory, base);
save_weights(net, buff); save_weights(net, buff);
} }
free_data(train); free_data(train);
@ -791,7 +791,7 @@ void predict_classifier(char *datacfg, char *cfgfile, char *weightfile, char *fi
int* indexes = (int*)calloc(top, sizeof(int)); int* indexes = (int*)calloc(top, sizeof(int));
char buff[256]; char buff[256];
char *input = buff; char *input = buff;
int size = net.w; //int size = net.w;
while(1){ while(1){
if(filename){ if(filename){
strncpy(input, filename, 256); strncpy(input, filename, 256);

2
src/compare.c
Unescape View File

@ -54,7 +54,7 @@ void train_compare(char *cfgfile, char *weightfile)
float loss = train_network(net, train); float loss = train_network(net, train);
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1; avg_loss = avg_loss*.9 + loss*.1;
printf("%.3f: %f, %f avg, %lf seconds, %d images\n", (float)*net.seen/N, loss, avg_loss, sec(clock()-time), *net.seen); printf("%.3f: %f, %f avg, %lf seconds, %ld images\n", (float)*net.seen/N, loss, avg_loss, sec(clock()-time), *net.seen);
free_data(train); free_data(train);
if(i%100 == 0){ if(i%100 == 0){
char buff[256]; char buff[256];

25
src/convolutional_layer.c
Unescape View File

@ -679,7 +679,7 @@ void float_to_bit(float *src, unsigned char *dst, size_t size) {
void bit_to_float(unsigned char *src, float *dst, size_t size, size_t filters, float *mean_arr) { void bit_to_float(unsigned char *src, float *dst, size_t size, size_t filters, float *mean_arr) {
memset(dst, 0, size *sizeof(float)); memset(dst, 0, size *sizeof(float));
size_t i, src_i, src_shift; size_t i;
for (i = 0; i < size; ++i) { for (i = 0; i < size; ++i) {
float mean_val = 1; float mean_val = 1;
@ -726,7 +726,7 @@ void binary_align_weights(convolutional_layer *l)
const int items_per_channel = l->size*l->size; const int items_per_channel = l->size*l->size;
for (i = 0; i < items_per_channel; ++i) for (i = 0; i < items_per_channel; ++i)
{ {
uint32_t val = 0; //uint32_t val = 0;
int c_pack; int c_pack;
for (c_pack = 0; c_pack < 32; ++c_pack) { for (c_pack = 0; c_pack < 32; ++c_pack) {
float src = l->binary_weights[fil*items_per_filter + (chan + c_pack)*items_per_channel + i]; float src = l->binary_weights[fil*items_per_filter + (chan + c_pack)*items_per_channel + i];
@ -749,8 +749,8 @@ void binary_align_weights(convolutional_layer *l)
//if (l->n >= 32) //if (l->n >= 32)
if(gpu_index >= 0) if(gpu_index >= 0)
{ {
int M = l->n; //int M = l->n;
int N = l->out_w*l->out_h; //int N = l->out_w*l->out_h;
//printf("\n M = %d, N = %d, M %% 8 = %d, N %% 8 = %d - weights \n", M, N, M % 8, N % 8); //printf("\n M = %d, N = %d, M %% 8 = %d, N %% 8 = %d - weights \n", M, N, M % 8, N % 8);
//printf("\n l.w = %d, l.c = %d, l.n = %d \n", l->w, l->c, l->n); //printf("\n l.w = %d, l.c = %d, l.n = %d \n", l->w, l->c, l->n);
for (i = 0; i < align_weights_size / 8; ++i) l->align_bit_weights[i] = ~(l->align_bit_weights[i]); for (i = 0; i < align_weights_size / 8; ++i) l->align_bit_weights[i] = ~(l->align_bit_weights[i]);
@ -806,7 +806,7 @@ size_t binary_transpose_align_input(int k, int n, float *b, char **t_bit_input,
size_t t_bit_input_size = t_intput_size / 8;// +1; size_t t_bit_input_size = t_intput_size / 8;// +1;
memset(*t_bit_input, 0, t_bit_input_size * sizeof(char)); memset(*t_bit_input, 0, t_bit_input_size * sizeof(char));
int src_size = k * bit_align; //int src_size = k * bit_align;
// b - [bit_align, k] - [l.bit_align, l.size*l.size*l.c] = src_size // b - [bit_align, k] - [l.bit_align, l.size*l.size*l.c] = src_size
// t_input - [bit_align, k] - [n', k] // t_input - [bit_align, k] - [n', k]
@ -862,8 +862,8 @@ void forward_convolutional_layer(convolutional_layer l, network_state state)
int ldb_align = l.lda_align; int ldb_align = l.lda_align;
size_t new_ldb = k + (ldb_align - k%ldb_align); // (k / 8 + 1) * 8; size_t new_ldb = k + (ldb_align - k%ldb_align); // (k / 8 + 1) * 8;
size_t t_intput_size = new_ldb * l.bit_align;// n; //size_t t_intput_size = new_ldb * l.bit_align;// n;
size_t t_bit_input_size = t_intput_size / 8;// +1; //size_t t_bit_input_size = t_intput_size / 8;// +1;
int re_packed_input_size = l.c * l.w * l.h; int re_packed_input_size = l.c * l.w * l.h;
memset(state.workspace, 0, re_packed_input_size * sizeof(float)); memset(state.workspace, 0, re_packed_input_size * sizeof(float));
@ -928,17 +928,18 @@ void forward_convolutional_layer(convolutional_layer l, network_state state)
//im2col_cpu_custom_align(state.input, l.c, l.h, l.w, l.size, l.stride, l.pad, b, l.bit_align); //im2col_cpu_custom_align(state.input, l.c, l.h, l.w, l.size, l.stride, l.pad, b, l.bit_align);
im2col_cpu_custom_bin(state.input, l.c, l.h, l.w, l.size, l.stride, l.pad, state.workspace, l.bit_align); im2col_cpu_custom_bin(state.input, l.c, l.h, l.w, l.size, l.stride, l.pad, state.workspace, l.bit_align);
size_t output_size = l.outputs; //size_t output_size = l.outputs;
//float *count_output = calloc(output_size, sizeof(float)); //float *count_output = calloc(output_size, sizeof(float));
//size_t bit_output_size = output_size / 8 + 1; //size_t bit_output_size = output_size / 8 + 1;
//char *bit_output = calloc(bit_output_size, sizeof(char)); //char *bit_output = calloc(bit_output_size, sizeof(char));
size_t intput_size = n * k; // (out_h*out_w) X (l.size*l.size*l.c) : after im2col() //size_t intput_size = n * k; // (out_h*out_w) X (l.size*l.size*l.c) : after im2col()
size_t bit_input_size = intput_size / 8 + 1; //size_t bit_input_size = intput_size / 8 + 1;
//char *bit_input = calloc(bit_input_size, sizeof(char)); //char *bit_input = calloc(bit_input_size, sizeof(char));
size_t weights_size = k * m; //l.size*l.size*l.c*l.n; //size_t weights_size = k * m; //l.size*l.size*l.c*l.n;
size_t bit_weights_size = weights_size / 8 + 1; //size_t bit_weights_size = weights_size / 8 + 1;
//char *bit_weights = calloc(bit_weights_size, sizeof(char)); //char *bit_weights = calloc(bit_weights_size, sizeof(char));
//float *mean_arr = calloc(l.n, sizeof(float)); //float *mean_arr = calloc(l.n, sizeof(float));

2
src/darknet.c
Unescape View File

@ -479,7 +479,7 @@ int main(int argc, char **argv)
float thresh = find_float_arg(argc, argv, "-thresh", .24); float thresh = find_float_arg(argc, argv, "-thresh", .24);
int ext_output = find_arg(argc, argv, "-ext_output"); int ext_output = find_arg(argc, argv, "-ext_output");
char *filename = (argc > 4) ? argv[4]: 0; char *filename = (argc > 4) ? argv[4]: 0;
test_detector("cfg/coco.data", argv[2], argv[3], filename, thresh, 0.5, 0, 1, 0, NULL); test_detector("cfg/coco.data", argv[2], argv[3], filename, thresh, 0.5, 0, ext_output, 0, NULL);
} else if (0 == strcmp(argv[1], "cifar")){ } else if (0 == strcmp(argv[1], "cifar")){
run_cifar(argc, argv); run_cifar(argc, argv);
} else if (0 == strcmp(argv[1], "go")){ } else if (0 == strcmp(argv[1], "go")){

2
src/darkunistd.h
Unescape View File

@ -21,7 +21,7 @@
These may be OR'd together. */ These may be OR'd together. */
#define R_OK 4 /* Test for read permission. */ #define R_OK 4 /* Test for read permission. */
#define W_OK 2 /* Test for write permission. */ #define W_OK 2 /* Test for write permission. */
#define X_OK R_OK /* execute permission - unsupported in Windows, \ #define X_OK R_OK /* execute permission - unsupported in Windows, */
#define F_OK 0 /* Test for existence. */ #define F_OK 0 /* Test for existence. */
#define access _access #define access _access

8
src/data.c
Unescape View File

@ -790,8 +790,8 @@ data load_data_detection(int n, char **paths, int m, int w, int h, int c, int bo
d.X.vals = (float**)calloc(d.X.rows, sizeof(float*)); d.X.vals = (float**)calloc(d.X.rows, sizeof(float*));
d.X.cols = h*w*c; d.X.cols = h*w*c;
float r1, r2, r3, r4; float r1 = 0, r2 = 0, r3 = 0, r4 = 0;
float dhue, dsat, dexp, flip; float dhue = 0, dsat = 0, dexp = 0, flip = 0;
int augmentation_calculated = 0; int augmentation_calculated = 0;
d.y = make_matrix(n, 5*boxes); d.y = make_matrix(n, 5*boxes);
@ -890,8 +890,8 @@ data load_data_detection(int n, char **paths, int m, int w, int h, int c, int bo
d.X.vals = (float**)calloc(d.X.rows, sizeof(float*)); d.X.vals = (float**)calloc(d.X.rows, sizeof(float*));
d.X.cols = h*w*c; d.X.cols = h*w*c;
float r1, r2, r3, r4; float r1 = 0, r2 = 0, r3 = 0, r4 = 0;
float dhue, dsat, dexp, flip; float dhue = 0, dsat = 0, dexp = 0, flip = 0;
int augmentation_calculated = 0; int augmentation_calculated = 0;
d.y = make_matrix(n, 5 * boxes); d.y = make_matrix(n, 5 * boxes);

21
src/detector.c
Unescape View File

@ -326,7 +326,7 @@ void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, i
if (ngpus != 1) sync_nets(nets, ngpus, 0); if (ngpus != 1) sync_nets(nets, ngpus, 0);
#endif #endif
char buff[256]; char buff[256];
sprintf(buff, "%s/%s_last.weights", backup_directory, base, i); sprintf(buff, "%s/%s_last.weights", backup_directory, base);
save_weights(net, buff); save_weights(net, buff);
} }
free_data(train); free_data(train);
@ -594,7 +594,7 @@ void validate_detector_recall(char *datacfg, char *cfgfile, char *weightfile)
list *plist = get_paths(valid_images); list *plist = get_paths(valid_images);
char **paths = (char **)list_to_array(plist); char **paths = (char **)list_to_array(plist);
layer l = net.layers[net.n - 1]; //layer l = net.layers[net.n - 1];
int j, k; int j, k;
@ -681,16 +681,16 @@ float validate_detector_map(char *datacfg, char *cfgfile, char *weightfile, floa
char *difficult_valid_images = option_find_str(options, "difficult", NULL); char *difficult_valid_images = option_find_str(options, "difficult", NULL);
char *name_list = option_find_str(options, "names", "data/names.list"); char *name_list = option_find_str(options, "names", "data/names.list");
char **names = get_labels(name_list); char **names = get_labels(name_list);
char *mapf = option_find_str(options, "map", 0); //char *mapf = option_find_str(options, "map", 0);
int *map = 0; //int *map = 0;
if (mapf) map = read_map(mapf); //if (mapf) map = read_map(mapf);
FILE* reinforcement_fd = NULL; FILE* reinforcement_fd = NULL;
network net; network net;
int initial_batch; //int initial_batch;
if (existing_net) { if (existing_net) {
char *train_images = option_find_str(options, "train", "data/train.txt"); char *train_images = option_find_str(options, "train", "data/train.txt");
char *valid_images = option_find_str(options, "valid", train_images); valid_images = option_find_str(options, "valid", train_images);
net = *existing_net; net = *existing_net;
} }
else { else {
@ -1131,8 +1131,8 @@ void calc_anchors(char *datacfg, int num_of_clusters, int width, int height, int
printf("\n"); printf("\n");
for (i = 0; i < number_of_boxes; ++i) { for (i = 0; i < number_of_boxes; ++i) {
float w = boxes_data.vals[i][0] = rel_width_height_array[i * 2]; boxes_data.vals[i][0] = rel_width_height_array[i * 2];
float h = boxes_data.vals[i][1] = rel_width_height_array[i * 2 + 1]; boxes_data.vals[i][1] = rel_width_height_array[i * 2 + 1];
//if (w > 410 || h > 410) printf("i:%d, w = %f, h = %f \n", i, w, h); //if (w > 410 || h > 410) printf("i:%d, w = %f, h = %f \n", i, w, h);
} }
@ -1170,7 +1170,7 @@ void calc_anchors(char *datacfg, int num_of_clusters, int width, int height, int
float anchor_w = anchors_data.centers.vals[cluster_idx][0]; //centers->data.fl[cluster_idx * 2]; float anchor_w = anchors_data.centers.vals[cluster_idx][0]; //centers->data.fl[cluster_idx * 2];
float anchor_h = anchors_data.centers.vals[cluster_idx][1]; //centers->data.fl[cluster_idx * 2 + 1]; float anchor_h = anchors_data.centers.vals[cluster_idx][1]; //centers->data.fl[cluster_idx * 2 + 1];
if (best_iou > 1 || best_iou < 0) { // || box_w > width || box_h > height) { if (best_iou > 1 || best_iou < 0) { // || box_w > width || box_h > height) {
printf(" Wrong label: i = %d, box_w = %d, box_h = %d, anchor_w = %d, anchor_h = %d, iou = %f \n", printf(" Wrong label: i = %d, box_w = %f, box_h = %f, anchor_w = %f, anchor_h = %f, iou = %f \n",
i, box_w, box_h, anchor_w, anchor_h, best_iou); i, box_w, box_h, anchor_w, anchor_h, best_iou);
} }
else avg_iou += best_iou; else avg_iou += best_iou;
@ -1287,7 +1287,6 @@ void test_detector(char *datacfg, char *cfgfile, char *weightfile, char *filenam
if (net.layers[net.n - 1].classes > names_size) getchar(); if (net.layers[net.n - 1].classes > names_size) getchar();
} }
srand(2222222); srand(2222222);
double time;
char buff[256]; char buff[256];
char *input = buff; char *input = buff;
char *json_buf = NULL; char *json_buf = NULL;

2
src/dice.c
Unescape View File

@ -33,7 +33,7 @@ void train_dice(char *cfgfile, char *weightfile)
float loss = train_network(net, train); float loss = train_network(net, train);
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1; 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, %ld images\n", i, loss, avg_loss, sec(clock()-time), *net.seen);
free_data(train); free_data(train);
if((i % 100) == 0) net.learning_rate *= .1; if((i % 100) == 0) net.learning_rate *= .1;
if(i%100==0){ if(i%100==0){

28
src/gemm.c
Unescape View File

@ -321,7 +321,7 @@ void gemm_nn_custom_bin_mean_transposed(int M, int N, int K, float ALPHA_UNUSED,
// is not used // is not used
void transpose_32x32_bits_my(uint32_t *A, uint32_t *B, int lda, int ldb) void transpose_32x32_bits_my(uint32_t *A, uint32_t *B, int lda, int ldb)
{ {
unsigned x, y, t; unsigned int x, y;
for (y = 0; y < 32; ++y) { for (y = 0; y < 32; ++y) {
for (x = 0; x < 32; ++x) { for (x = 0; x < 32; ++x) {
if (A[y * lda] & (1 << x)) B[x * ldb] |= (uint32_t)1 << y; if (A[y * lda] & (1 << x)) B[x * ldb] |= (uint32_t)1 << y;
@ -400,7 +400,7 @@ void transpose_32x32_bits_reversed_diagonale(uint32_t *A, uint32_t *B, int m, in
void transpose_8x8_bits_my(unsigned char *A, unsigned char *B, int lda, int ldb) void transpose_8x8_bits_my(unsigned char *A, unsigned char *B, int lda, int ldb)
{ {
unsigned x, y, t; unsigned x, y;
for (y = 0; y < 8; ++y) { for (y = 0; y < 8; ++y) {
for (x = 0; x < 8; ++x) { for (x = 0; x < 8; ++x) {
if (A[y * lda] & (1 << x)) B[x * ldb] |= 1 << y; if (A[y * lda] & (1 << x)) B[x * ldb] |= 1 << y;
@ -755,7 +755,7 @@ void gemm_nn_fast(int M, int N, int K, float ALPHA,
for (i = 0; i < (M / TILE_M)*TILE_M; i += TILE_M) for (i = 0; i < (M / TILE_M)*TILE_M; i += TILE_M)
{ {
int j, k; int j, k;
int i_d, j_d, k_d; int i_d, k_d;
for (k = 0; k < (K / TILE_K)*TILE_K; k += TILE_K) for (k = 0; k < (K / TILE_K)*TILE_K; k += TILE_K)
{ {
@ -768,8 +768,8 @@ void gemm_nn_fast(int M, int N, int K, float ALPHA,
__m256 result256; __m256 result256;
__m256 a256_0, b256_0; // AVX __m256 a256_0, b256_0; // AVX
__m256 a256_1, b256_1; // AVX __m256 a256_1, b256_1; // AVX
__m256 a256_2, b256_2; // AVX __m256 a256_2;// , b256_2; // AVX
__m256 a256_3, b256_3; // AVX __m256 a256_3;// , b256_3; // AVX
__m256 c256_0, c256_1, c256_2, c256_3; __m256 c256_0, c256_1, c256_2, c256_3;
__m256 c256_4, c256_5, c256_6, c256_7; __m256 c256_4, c256_5, c256_6, c256_7;
@ -943,8 +943,8 @@ void gemm_nn_bin_32bit_packed(int M, int N, int K, float ALPHA,
void convolution_2d_old(int w, int h, int ksize, int n, int c, int pad, int stride, void convolution_2d_old(int w, int h, int ksize, int n, int c, int pad, int stride,
float *weights, float *input, float *output) float *weights, float *input, float *output)
{ {
const int out_h = (h + 2 * pad - ksize) / stride + 1; // output_height=input_height for stride=1 and pad=1 //const int out_h = (h + 2 * pad - ksize) / stride + 1; // output_height=input_height for stride=1 and pad=1
const int out_w = (w + 2 * pad - ksize) / stride + 1; // output_width=input_width for stride=1 and pad=1 //const int out_w = (w + 2 * pad - ksize) / stride + 1; // output_width=input_width for stride=1 and pad=1
int fil; int fil;
// filter index // filter index
@ -991,8 +991,8 @@ void convolution_2d_old(int w, int h, int ksize, int n, int c, int pad, int stri
void convolution_2d(int w, int h, int ksize, int n, int c, int pad, int stride, void convolution_2d(int w, int h, int ksize, int n, int c, int pad, int stride,
float *weights, float *input, float *output, float *mean) float *weights, float *input, float *output, float *mean)
{ {
const int out_h = (h + 2 * pad - ksize) / stride + 1; // output_height=input_height for stride=1 and pad=1 //const int out_h = (h + 2 * pad - ksize) / stride + 1; // output_height=input_height for stride=1 and pad=1
const int out_w = (w + 2 * pad - ksize) / stride + 1; // output_width=input_width for stride=1 and pad=1 //const int out_w = (w + 2 * pad - ksize) / stride + 1; // output_width=input_width for stride=1 and pad=1
int i; int i;
#if defined(_OPENMP) #if defined(_OPENMP)
@ -1203,7 +1203,7 @@ void gemm_nn_custom_bin_mean_transposed(int M, int N, int K, float ALPHA_UNUSED,
float mean_val_0 = mean_arr[i + 0]; float mean_val_0 = mean_arr[i + 0];
float mean_val_1 = mean_arr[i + 1]; float mean_val_1 = mean_arr[i + 1];
int j, k; int j, k;
__m256i all_1 = _mm256_set1_epi8(255); //__m256i all_1 = _mm256_set1_epi8(255);
//for (j = 0; j < N; ++j) //for (j = 0; j < N; ++j)
for (j = 0; j < (N/2)*2; j += 2) for (j = 0; j < (N/2)*2; j += 2)
@ -1770,7 +1770,7 @@ void float_to_bit(float *src, unsigned char *dst, size_t size)
memset(dst, 0, dst_size); memset(dst, 0, dst_size);
size_t i; size_t i;
__m256i all256_sing1 = _mm256_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000); //__m256i all256_sing1 = _mm256_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000);
__m256 float_zero256 = _mm256_set1_ps(0.0); __m256 float_zero256 = _mm256_set1_ps(0.0);
for (i = 0; i < size; i+=8) for (i = 0; i < size; i+=8)
@ -1881,8 +1881,8 @@ void forward_maxpool_layer_avx(float *src, float *dst, int *indexes, int size, i
else if (size == 2 && stride == 2 && is_avx() == 1) { else if (size == 2 && stride == 2 && is_avx() == 1) {
for (j = 0; j < out_w - 4; j += 4) { for (j = 0; j < out_w - 4; j += 4) {
int out_index = j + out_w*(i + out_h*(k + c*b)); int out_index = j + out_w*(i + out_h*(k + c*b));
float max = -FLT_MAX; //float max = -FLT_MAX;
int max_i = -1; //int max_i = -1;
__m128 max128 = _mm_set1_ps(-FLT_MAX); __m128 max128 = _mm_set1_ps(-FLT_MAX);
for (n = 0; n < size; ++n) { for (n = 0; n < size; ++n) {
@ -2513,7 +2513,7 @@ void convolution_repacked(uint32_t *packed_input, uint32_t *packed_weights, floa
#pragma omp parallel for #pragma omp parallel for
for (fil = 0; fil < n; ++fil) { for (fil = 0; fil < n; ++fil) {
float mean_val = mean_arr[fil]; float mean_val = mean_arr[fil];
int chan, c_pack, y, x, f_y, f_x; int chan, y, x, f_y, f_x; // c_pack
// channel index // channel index
for (chan = 0; chan < c / 32; ++chan) for (chan = 0; chan < c / 32; ++chan)
//for (chan = 0; chan < l.c; chan += 32) //for (chan = 0; chan < l.c; chan += 32)

2
src/go.c
Unescape View File

@ -144,7 +144,7 @@ void train_go(char *cfgfile, char *weightfile)
float loss = train_network_datum(net, board, move) / net.batch; float loss = train_network_datum(net, board, move) / net.batch;
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.95 + loss*.05; avg_loss = avg_loss*.95 + loss*.05;
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); printf("%d, %.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){ if(*net.seen/N > epoch){
epoch = *net.seen/N; epoch = *net.seen/N;
char buff[256]; char buff[256];

2
src/http_stream.cpp
Unescape View File

@ -181,7 +181,7 @@ public:
if (::select(maxfd + 1, &rread, NULL, NULL, &select_timeout) <= 0) if (::select(maxfd + 1, &rread, NULL, NULL, &select_timeout) <= 0)
return true; // nothing broken, there's just noone listening return true; // nothing broken, there's just noone listening
size_t outlen = strlen(outputbuf); int outlen = static_cast<int>(strlen(outputbuf));
#ifdef _WIN32 #ifdef _WIN32
for (unsigned i = 0; i<rread.fd_count; i++) for (unsigned i = 0; i<rread.fd_count; i++)

3
src/matrix.c
Unescape View File

@ -257,7 +257,7 @@ void kmeans_maximization(matrix data, int *assignments, matrix centers)
void random_centers(matrix data, matrix centers) { void random_centers(matrix data, matrix centers) {
int i, j; int i;
int *s = sample(data.rows); int *s = sample(data.rows);
for (i = 0; i < centers.rows; ++i) { for (i = 0; i < centers.rows; ++i) {
copy(data.vals[s[i]], centers.vals[i], data.cols); copy(data.vals[s[i]], centers.vals[i], data.cols);
@ -281,7 +281,6 @@ int *sample(int n)
float dist(float *x, float *y, int n) float dist(float *x, float *y, int n)
{ {
int i;
//printf(" x0 = %f, x1 = %f, y0 = %f, y1 = %f \n", x[0], x[1], y[0], y[1]); //printf(" x0 = %f, x1 = %f, y0 = %f, y1 = %f \n", x[0], x[1], y[0], y[1]);
float mw = (x[0] < y[0]) ? x[0] : y[0]; float mw = (x[0] < y[0]) ? x[0] : y[0];
float mh = (x[1] < y[1]) ? x[1] : y[1]; float mh = (x[1] < y[1]) ? x[1] : y[1];

6
src/network.c
Unescape View File

@ -488,8 +488,8 @@ int resize_network(network *net, int w, int h)
h = l.out_h; h = l.out_h;
if(l.type == AVGPOOL) break; if(l.type == AVGPOOL) break;
} }
const int size = get_network_input_size(*net) * net->batch;
#ifdef GPU #ifdef GPU
const int size = get_network_input_size(*net) * net->batch;
if(gpu_index >= 0){ if(gpu_index >= 0){
printf(" try to allocate additional workspace_size = %1.2f MB \n", (float)workspace_size / 1000000); printf(" try to allocate additional workspace_size = %1.2f MB \n", (float)workspace_size / 1000000);
net->workspace = cuda_make_array(0, workspace_size/sizeof(float) + 1); net->workspace = cuda_make_array(0, workspace_size/sizeof(float) + 1);
@ -728,10 +728,10 @@ char *detection_to_json(detection *dets, int nboxes, int classes, char **names,
char *send_buf = (char *)calloc(1024, sizeof(char)); char *send_buf = (char *)calloc(1024, sizeof(char));
if (filename) { if (filename) {
sprintf(send_buf, "{\n \"frame_id\":%d, \n \"filename\":\"%s\", \n \"objects\": [ \n", frame_id, filename); sprintf(send_buf, "{\n \"frame_id\":%lld, \n \"filename\":\"%s\", \n \"objects\": [ \n", frame_id, filename);
} }
else { else {
sprintf(send_buf, "{\n \"frame_id\":%d, \n \"objects\": [ \n", frame_id); sprintf(send_buf, "{\n \"frame_id\":%lld, \n \"objects\": [ \n", frame_id);
} }
int i, j; int i, j;

2
src/rnn.c
Unescape View File

@ -163,7 +163,7 @@ void train_char_rnn(char *cfgfile, char *weightfile, char *filename, int clear,
int i = (*net.seen)/net.batch; int i = (*net.seen)/net.batch;
int streams = batch/steps; int streams = batch/steps;
printf("\n batch = %d, steps = %d, streams = %d, subdivisions = %d, text_size = %d \n", batch, steps, streams, net.subdivisions, size); printf("\n batch = %d, steps = %d, streams = %d, subdivisions = %d, text_size = %ld \n", batch, steps, streams, net.subdivisions, size);
printf(" global_batch = %d \n", batch*net.subdivisions); printf(" global_batch = %d \n", batch*net.subdivisions);
size_t* offsets = (size_t*)calloc(streams, sizeof(size_t)); size_t* offsets = (size_t*)calloc(streams, sizeof(size_t));
int j; int j;

2
src/tag.c
Unescape View File

@ -64,7 +64,7 @@ void train_tag(char *cfgfile, char *weightfile, int clear)
float loss = train_network(net, train); float loss = train_network(net, train);
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1; 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); printf("%d, %.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); free_data(train);
if(*net.seen/N > epoch){ if(*net.seen/N > epoch){
epoch = *net.seen/N; epoch = *net.seen/N;

17
src/utils.c
Unescape View File

@ -254,7 +254,7 @@ void replace_image_to_label(const char* input_path, char* output_path)
find_replace_extension(output_path, ".PPM", ".txt", output_path); find_replace_extension(output_path, ".PPM", ".txt", output_path);
find_replace_extension(output_path, ".tiff", ".txt", output_path); find_replace_extension(output_path, ".tiff", ".txt", output_path);
find_replace_extension(output_path, ".TIFF", ".txt", output_path); find_replace_extension(output_path, ".TIFF", ".txt", output_path);
// Check file ends with txt: // Check file ends with txt:
if(strlen(output_path) > 4) { if(strlen(output_path) > 4) {
char *output_path_ext = output_path + strlen(output_path) - 4; char *output_path_ext = output_path + strlen(output_path) - 4;
@ -768,13 +768,12 @@ float rand_uniform(float min, float max)
max = swap; max = swap;
} }
if (RAND_MAX < 65536) { #if (RAND_MAX < 65536)
int rnd = rand()*(RAND_MAX + 1) + rand(); int rnd = rand()*(RAND_MAX + 1) + rand();
return ((float)rnd / (RAND_MAX*RAND_MAX) * (max - min)) + min; return ((float)rnd / (RAND_MAX*RAND_MAX) * (max - min)) + min;
} #else
else {
return ((float)rand() / RAND_MAX * (max - min)) + min; return ((float)rand() / RAND_MAX * (max - min)) + min;
} #endif
//return (random_float() * (max - min)) + min; //return (random_float() * (max - min)) + min;
} }
@ -802,12 +801,12 @@ unsigned int random_gen()
unsigned int rnd = 0; unsigned int rnd = 0;
#ifdef WIN32 #ifdef WIN32
rand_s(&rnd); rand_s(&rnd);
#else #else // WIN32
rnd = rand(); rnd = rand();
if (RAND_MAX < 65536) { #if (RAND_MAX < 65536)
rnd = rand()*(RAND_MAX + 1) + rnd; rnd = rand()*(RAND_MAX + 1) + rnd;
} #endif //(RAND_MAX < 65536)
#endif #endif // WIN32
return rnd; return rnd;
} }

2
src/writing.c
Unescape View File

@ -69,7 +69,7 @@ void train_writing(char *cfgfile, char *weightfile)
if(avg_loss == -1) avg_loss = loss; if(avg_loss == -1) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1; 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); printf("%d, %.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); free_data(train);
if(get_current_batch(net)%100 == 0){ if(get_current_batch(net)%100 == 0){
char buff[256]; char buff[256];

10
src/yolo_v2_class.cpp
Unescape View File

@ -104,8 +104,8 @@ struct detector_gpu_t {
LIB_API Detector::Detector(std::string cfg_filename, std::string weight_filename, int gpu_id) : cur_gpu_id(gpu_id) LIB_API Detector::Detector(std::string cfg_filename, std::string weight_filename, int gpu_id) : cur_gpu_id(gpu_id)
{ {
wait_stream = 0; wait_stream = 0;
int old_gpu_index;
#ifdef GPU #ifdef GPU
int old_gpu_index;
check_cuda( cudaGetDevice(&old_gpu_index) ); check_cuda( cudaGetDevice(&old_gpu_index) );
#endif #endif
@ -151,7 +151,7 @@ LIB_API Detector::Detector(std::string cfg_filename, std::string weight_filename
LIB_API Detector::~Detector() LIB_API Detector::~Detector()
{ {
detector_gpu_t &detector_gpu = *static_cast<detector_gpu_t *>(detector_gpu_ptr.get()); detector_gpu_t &detector_gpu = *static_cast<detector_gpu_t *>(detector_gpu_ptr.get());
layer l = detector_gpu.net.layers[detector_gpu.net.n - 1]; //layer l = detector_gpu.net.layers[detector_gpu.net.n - 1];
free(detector_gpu.track_id); free(detector_gpu.track_id);
@ -159,8 +159,8 @@ LIB_API Detector::~Detector()
for (int j = 0; j < NFRAMES; ++j) free(detector_gpu.predictions[j]); for (int j = 0; j < NFRAMES; ++j) free(detector_gpu.predictions[j]);
for (int j = 0; j < NFRAMES; ++j) if (detector_gpu.images[j].data) free(detector_gpu.images[j].data); for (int j = 0; j < NFRAMES; ++j) if (detector_gpu.images[j].data) free(detector_gpu.images[j].data);
int old_gpu_index;
#ifdef GPU #ifdef GPU
int old_gpu_index;
cudaGetDevice(&old_gpu_index); cudaGetDevice(&old_gpu_index);
cuda_set_device(detector_gpu.net.gpu_index); cuda_set_device(detector_gpu.net.gpu_index);
#endif #endif
@ -241,8 +241,8 @@ LIB_API std::vector<bbox_t> Detector::detect(image_t img, float thresh, bool use
{ {
detector_gpu_t &detector_gpu = *static_cast<detector_gpu_t *>(detector_gpu_ptr.get()); detector_gpu_t &detector_gpu = *static_cast<detector_gpu_t *>(detector_gpu_ptr.get());
network &net = detector_gpu.net; network &net = detector_gpu.net;
int old_gpu_index;
#ifdef GPU #ifdef GPU
int old_gpu_index;
cudaGetDevice(&old_gpu_index); cudaGetDevice(&old_gpu_index);
if(cur_gpu_id != old_gpu_index) if(cur_gpu_id != old_gpu_index)
cudaSetDevice(net.gpu_index); cudaSetDevice(net.gpu_index);
@ -289,7 +289,7 @@ LIB_API std::vector<bbox_t> Detector::detect(image_t img, float thresh, bool use
std::vector<bbox_t> bbox_vec; std::vector<bbox_t> bbox_vec;
for (size_t i = 0; i < nboxes; ++i) { for (int i = 0; i < nboxes; ++i) {
box b = dets[i].bbox; box b = dets[i].bbox;
int const obj_id = max_index(dets[i].prob, l.classes); int const obj_id = max_index(dets[i].prob, l.classes);
float const prob = dets[i].prob[obj_id]; float const prob = dets[i].prob[obj_id];

Write Preview
Loading…
Cancel
Save