mirror/darknet
1
0
Fork 0
mirror of https://github.com/pjreddie/darknet.git synced 2023-08-10 21:13:14 +03:00
Code Issues Projects Releases Wiki Activity

stable

Browse Source
This commit is contained in:
Joseph Redmon 2015-09-09 12:48:40 -07:00
parent b5936b499a
commit 393dc8eb6f
10 changed files with 415 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Makefile
View File

@ -34,7 +34,7 @@ CFLAGS+= -DGPU
LDFLAGS+= -L/usr/local/cuda/lib64 -lcuda -lcudart -lcublas -lcurand
endif
OBJ=gemm.o utils.o cuda.o deconvolutional_layer.o convolutional_layer.o list.o image.o activations.o im2col.o col2im.o blas.o crop_layer.o dropout_layer.o maxpool_layer.o softmax_layer.o data.o matrix.o network.o connected_layer.o cost_layer.o parser.o option_list.o darknet.o detection_layer.o imagenet.o captcha.o route_layer.o writing.o box.o nightmare.o normalization_layer.o avgpool_layer.o coco.o dice.o yolo.o region_layer.o layer.o compare.o
OBJ=gemm.o utils.o cuda.o deconvolutional_layer.o convolutional_layer.o list.o image.o activations.o im2col.o col2im.o blas.o crop_layer.o dropout_layer.o maxpool_layer.o softmax_layer.o data.o matrix.o network.o connected_layer.o cost_layer.o parser.o option_list.o darknet.o detection_layer.o imagenet.o captcha.o route_layer.o writing.o box.o nightmare.o normalization_layer.o avgpool_layer.o coco.o dice.o yolo.o region_layer.o layer.o compare.o yoloplus.o
ifeq ($(GPU), 1)
OBJ+=convolutional_kernels.o deconvolutional_kernels.o activation_kernels.o im2col_kernels.o col2im_kernels.o blas_kernels.o crop_layer_kernels.o dropout_layer_kernels.o maxpool_layer_kernels.o softmax_layer_kernels.o network_kernels.o avgpool_layer_kernels.o
endif

14
cfg/darknet.cfg
View File

@ -27,7 +27,7 @@ pad=1
activation=leaky
[maxpool]
size=3
size=2
stride=2
[convolutional]
@ -38,7 +38,7 @@ pad=1
activation=leaky
[maxpool]
size=3
size=2
stride=2
[convolutional]
@ -49,7 +49,7 @@ pad=1
activation=leaky
[maxpool]
size=3
size=2
stride=2
[convolutional]
@ -60,7 +60,7 @@ pad=1
activation=leaky
[maxpool]
size=3
size=2
stride=2
[convolutional]
@ -71,7 +71,7 @@ pad=1
activation=leaky
[maxpool]
size=3
size=2
stride=2
[convolutional]
@ -82,7 +82,7 @@ pad=1
activation=leaky
[maxpool]
size=3
size=2
stride=2
[convolutional]
@ -99,7 +99,7 @@ probability=.5
[connected]
output=1000
activation=linear
activation=leaky
[softmax]

7
cfg/yolo.cfg
View File

@ -4,10 +4,15 @@ subdivisions=64
height=448
width=448
channels=3
learning_rate=0.01
learning_rate=0.001
momentum=0.9
decay=0.0005
policy=steps
steps=50, 5000
scales=10, .1
max_batches = 8000
[crop]
crop_width=448
crop_height=448

3
src/darknet.c
View File

@ -13,6 +13,7 @@
extern void run_imagenet(int argc, char **argv);
extern void run_yolo(int argc, char **argv);
extern void run_yoloplus(int argc, char **argv);
extern void run_coco(int argc, char **argv);
extern void run_writing(int argc, char **argv);
extern void run_captcha(int argc, char **argv);
@ -178,6 +179,8 @@ int main(int argc, char **argv)
average(argc, argv);
} else if (0 == strcmp(argv[1], "yolo")){
run_yolo(argc, argv);
} else if (0 == strcmp(argv[1], "yoloplus")){
run_yoloplus(argc, argv);
} else if (0 == strcmp(argv[1], "coco")){
run_coco(argc, argv);
} else if (0 == strcmp(argv[1], "compare")){

17
src/detection_layer.c
View File

@ -85,11 +85,12 @@ void forward_detection_layer(const detection_layer l, network_state state)
int size = get_detection_layer_output_size(l) * l.batch;
memset(l.delta, 0, size * sizeof(float));
for (i = 0; i < l.batch*locations; ++i) {
int classes = l.objectness+l.classes;
int classes = (l.objectness || l.background)+l.classes;
int offset = i*(classes+l.coords);
for (j = offset; j < offset+classes; ++j) {
*(l.cost) += pow(state.truth[j] - l.output[j], 2);
l.delta[j] = state.truth[j] - l.output[j];
if(l.background && j == offset) l.delta[j] *= .1;
}
box truth;
@ -115,9 +116,15 @@ void forward_detection_layer(const detection_layer l, network_state state)
l.delta[j+2] = 4 * (state.truth[j+2] - l.output[j+2]);
l.delta[j+3] = 4 * (state.truth[j+3] - l.output[j+3]);
if(l.rescore){
for (j = offset; j < offset+classes; ++j) {
if(state.truth[j]) state.truth[j] = iou;
l.delta[j] = state.truth[j] - l.output[j];
if(l.objectness){
state.truth[offset] = iou;
l.delta[offset] = state.truth[offset] - l.output[offset];
}
else{
for (j = offset; j < offset+classes; ++j) {
if(state.truth[j]) state.truth[j] = iou;
l.delta[j] = state.truth[j] - l.output[j];
}
}
}
}
@ -145,7 +152,7 @@ void backward_detection_layer(const detection_layer l, network_state state)
if (l.objectness) {
}else if (l.background) gradient_array(l.output + out_i, l.coords, LOGISTIC, l.delta + out_i);
for(j = 0; j < l.coords; ++j){
for (j = 0; j < l.coords; ++j){
state.delta[in_i++] += l.delta[out_i++];
}
if(l.joint) state.delta[in_i-l.coords-l.classes-l.joint] += latent_delta;

13
src/network.c
View File

@ -29,15 +29,26 @@ int get_current_batch(network net)
float get_current_rate(network net)
{
int batch_num = get_current_batch(net);
int i;
float rate;
switch (net.policy) {
case CONSTANT:
return net.learning_rate;
case STEP:
return net.learning_rate * pow(net.gamma, batch_num/net.step);
return net.learning_rate * pow(net.scale, batch_num/net.step);
case STEPS:
rate = net.learning_rate;
for(i = 0; i < net.num_steps; ++i){
if(net.steps[i] > batch_num) return rate;
rate *= net.scales[i];
}
return rate;
case EXP:
return net.learning_rate * pow(net.gamma, batch_num);
case POLY:
return net.learning_rate * pow(1 - (float)batch_num / net.max_batches, net.power);
case SIG:
return net.learning_rate * (1/(1+exp(net.gamma*(batch_num - net.step))));
default:
fprintf(stderr, "Policy is weird!\n");
return net.learning_rate;

6
src/network.h
View File

@ -8,7 +8,7 @@
#include "data.h"
typedef enum {
CONSTANT, STEP, EXP, POLY
CONSTANT, STEP, EXP, POLY, STEPS, SIG
} learning_rate_policy;
typedef struct {
@ -25,9 +25,13 @@ typedef struct {
float learning_rate;
float gamma;
float scale;
float power;
int step;
int max_batches;
float *scales;
int *steps;
int num_steps;
int inputs;
int h, w, c;

37
src/parser.c
View File

@ -169,7 +169,7 @@ detection_layer parse_detection(list *options, size_params params)
int rescore = option_find_int(options, "rescore", 0);
int joint = option_find_int(options, "joint", 0);
int objectness = option_find_int(options, "objectness", 0);
int background = 0;
int background = option_find_int(options, "background", 0);
detection_layer layer = make_detection_layer(params.batch, params.inputs, classes, coords, joint, rescore, background, objectness);
return layer;
}
@ -312,6 +312,8 @@ learning_rate_policy get_policy(char *s)
if (strcmp(s, "constant")==0) return CONSTANT;
if (strcmp(s, "step")==0) return STEP;
if (strcmp(s, "exp")==0) return EXP;
if (strcmp(s, "sigmoid")==0) return SIG;
if (strcmp(s, "steps")==0) return STEPS;
fprintf(stderr, "Couldn't find policy %s, going with constant\n", s);
return CONSTANT;
}
@ -337,9 +339,36 @@ void parse_net_options(list *options, network *net)
net->policy = get_policy(policy_s);
if(net->policy == STEP){
net->step = option_find_int(options, "step", 1);
net->gamma = option_find_float(options, "gamma", 1);
net->scale = option_find_float(options, "scale", 1);
} else if (net->policy == STEPS){
char *l = option_find(options, "steps");
char *p = option_find(options, "scales");
if(!l || !p) error("STEPS policy must have steps and scales in cfg file");
int len = strlen(l);
int n = 1;
int i;
for(i = 0; i < len; ++i){
if (l[i] == ',') ++n;
}
int *steps = calloc(n, sizeof(int));
float *scales = calloc(n, sizeof(float));
for(i = 0; i < n; ++i){
int step = atoi(l);
float scale = atof(p);
l = strchr(l, ',')+1;
p = strchr(p, ',')+1;
steps[i] = step;
scales[i] = scale;
}
net->scales = scales;
net->steps = steps;
net->num_steps = n;
} else if (net->policy == EXP){
net->gamma = option_find_float(options, "gamma", 1);
} else if (net->policy == SIG){
net->gamma = option_find_float(options, "gamma", 1);
net->step = option_find_int(options, "step", 1);
} else if (net->policy == POLY){
net->power = option_find_float(options, "power", 1);
}
@ -401,10 +430,10 @@ network parse_network_cfg(char *filename)
l = parse_dropout(options, params);
l.output = net.layers[count-1].output;
l.delta = net.layers[count-1].delta;
#ifdef GPU
#ifdef GPU
l.output_gpu = net.layers[count-1].output_gpu;
l.delta_gpu = net.layers[count-1].delta_gpu;
#endif
#endif
}else{
fprintf(stderr, "Type not recognized: %s\n", s->type);
}

18
src/yolo.c
View File

@ -66,7 +66,6 @@ void train_yolo(char *cfgfile, char *weightfile)
load_weights(&net, weightfile);
}
detection_layer layer = get_network_detection_layer(net);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = 128;
int i = *net.seen/imgs;
@ -75,10 +74,6 @@ void train_yolo(char *cfgfile, char *weightfile)
int N = plist->size;
paths = (char **)list_to_array(plist);
if(i*imgs > N*80){
net.layers[net.n-1].joint = 1;
net.layers[net.n-1].objectness = 0;
}
if(i*imgs > N*120){
net.layers[net.n-1].rescore = 1;
}
@ -102,7 +97,7 @@ void train_yolo(char *cfgfile, char *weightfile)
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
while(i*imgs < N*130){
while(get_current_batch(net) < net.max_batches){
i += 1;
time=clock();
pthread_join(load_thread, 0);
@ -115,19 +110,10 @@ void train_yolo(char *cfgfile, char *weightfile)
if (avg_loss < 0) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%d: %f, %f avg, %lf seconds, %d images, epoch: %f\n", i, loss, avg_loss, sec(clock()-time), i*imgs, ((float)i)*imgs/N);
if((i-1)*imgs <= N && i*imgs > N){
fprintf(stderr, "First stage done\n");
net.learning_rate *= 10;
char buff[256];
sprintf(buff, "%s/%s_first_stage.weights", backup_directory, base);
save_weights(net, buff);
}
printf("%d: %f, %f avg, %lf seconds, %f rate, %d images, epoch: %f\n", get_current_batch(net), loss, avg_loss, sec(clock()-time), get_current_rate(net), *net.seen, (float)*net.seen/N);
if((i-1)*imgs <= 80*N && i*imgs > N*80){
fprintf(stderr, "Second stage done.\n");
net.learning_rate *= .1;
char buff[256];
sprintf(buff, "%s/%s_second_stage.weights", backup_directory, base);
save_weights(net, buff);

334
src/yoloplus.c Normal file
View File

@ -0,0 +1,334 @@
#include "network.h"
#include "detection_layer.h"
#include "cost_layer.h"
#include "utils.h"
#include "parser.h"
#include "box.h"
#ifdef OPENCV
#include "opencv2/highgui/highgui_c.h"
#endif
char *voc_names[] = {"aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"};
void draw_yoloplus(image im, float *box, int side, int objectness, char *label, float thresh)
{
int classes = 20;
int elems = 4+classes+objectness;
int j;
int r, c;
for(r = 0; r < side; ++r){
for(c = 0; c < side; ++c){
j = (r*side + c) * elems;
float scale = 1;
if(objectness) scale = 1 - box[j++];
int class = max_index(box+j, classes);
if(scale * box[j+class] > thresh){
int width = sqrt(scale*box[j+class])*5 + 1;
printf("%f %s\n", scale * box[j+class], voc_names[class]);
float red = get_color(0,class,classes);
float green = get_color(1,class,classes);
float blue = get_color(2,class,classes);
j += classes;
float x = box[j+0];
float y = box[j+1];
x = (x+c)/side;
y = (y+r)/side;
float w = box[j+2]; //*maxwidth;
float h = box[j+3]; //*maxheight;
h = h*h;
w = w*w;
int left = (x-w/2)*im.w;
int right = (x+w/2)*im.w;
int top = (y-h/2)*im.h;
int bot = (y+h/2)*im.h;
draw_box_width(im, left, top, right, bot, width, red, green, blue);
}
}
}
show_image(im, label);
}
void train_yoloplus(char *cfgfile, char *weightfile)
{
char *train_images = "/home/pjreddie/data/voc/test/train.txt";
char *backup_directory = "/home/pjreddie/backup/";
srand(time(0));
data_seed = time(0);
char *base = basecfg(cfgfile);
printf("%s\n", base);
float avg_loss = -1;
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
detection_layer layer = get_network_detection_layer(net);
int imgs = 128;
int i = *net.seen/imgs;
char **paths;
list *plist = get_paths(train_images);
int N = plist->size;
paths = (char **)list_to_array(plist);
if(i*imgs > N*120){
net.layers[net.n-1].rescore = 1;
}
data train, buffer;
int classes = layer.classes;
int background = layer.objectness;
int side = sqrt(get_detection_layer_locations(layer));
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.num_boxes = side;
args.background = background;
args.d = &buffer;
args.type = DETECTION_DATA;
pthread_t load_thread = load_data_in_thread(args);
clock_t time;
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));
time=clock();
float loss = train_network(net, train);
if (avg_loss < 0) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
printf("%d: %f, %f avg, %lf seconds, %f rate, %d images, epoch: %f\n", get_current_batch(net), loss, avg_loss, sec(clock()-time), get_current_rate(net), *net.seen, (float)*net.seen/N);
if((i-1)*imgs <= 80*N && i*imgs > N*80){
fprintf(stderr, "Second stage done.\n");
char buff[256];
sprintf(buff, "%s/%s_second_stage.weights", backup_directory, base);
save_weights(net, buff);
net.layers[net.n-1].joint = 1;
net.layers[net.n-1].objectness = 0;
background = 0;
pthread_join(load_thread, 0);
free_data(buffer);
args.background = background;
load_thread = load_data_in_thread(args);
}
if((i-1)*imgs <= 120*N && i*imgs > N*120){
fprintf(stderr, "Third stage done.\n");
char buff[256];
sprintf(buff, "%s/%s_final.weights", backup_directory, base);
net.layers[net.n-1].rescore = 1;
save_weights(net, buff);
}
if(i%1000==0){
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
save_weights(net, buff);
}
free_data(train);
}
char buff[256];
sprintf(buff, "%s/%s_rescore.weights", backup_directory, base);
save_weights(net, buff);
}
void convert_yoloplus_detections(float *predictions, int classes, int objectness, int background, int num_boxes, int w, int h, float thresh, float **probs, box *boxes)
{
int i,j;
int per_box = 4+classes+(background || objectness);
for (i = 0; i < num_boxes*num_boxes; ++i){
float scale = 1;
if(objectness) scale = 1-predictions[i*per_box];
int offset = i*per_box+(background||objectness);
for(j = 0; j < classes; ++j){
float prob = scale*predictions[offset+j];
probs[i][j] = (prob > thresh) ? prob : 0;
}
int row = i / num_boxes;
int col = i % num_boxes;
offset += classes;
boxes[i].x = (predictions[offset + 0] + col) / num_boxes * w;
boxes[i].y = (predictions[offset + 1] + row) / num_boxes * h;
boxes[i].w = pow(predictions[offset + 2], 2) * w;
boxes[i].h = pow(predictions[offset + 3], 2) * h;
}
}
void print_yoloplus_detections(FILE **fps, char *id, box *boxes, float **probs, int num_boxes, int classes, int w, int h)
{
int i, j;
for(i = 0; i < num_boxes*num_boxes; ++i){
float xmin = boxes[i].x - boxes[i].w/2.;
float xmax = boxes[i].x + boxes[i].w/2.;
float ymin = boxes[i].y - boxes[i].h/2.;
float ymax = boxes[i].y + boxes[i].h/2.;
if (xmin < 0) xmin = 0;
if (ymin < 0) ymin = 0;
if (xmax > w) xmax = w;
if (ymax > h) ymax = h;
for(j = 0; j < classes; ++j){
if (probs[i][j]) fprintf(fps[j], "%s %f %f %f %f %f\n", id, probs[i][j],
xmin, ymin, xmax, ymax);
}
}
}
void validate_yoloplus(char *cfgfile, char *weightfile)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
set_batch_network(&net, 1);
detection_layer layer = get_network_detection_layer(net);
fprintf(stderr, "Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
srand(time(0));
char *base = "results/comp4_det_test_";
list *plist = get_paths("/home/pjreddie/data/voc/test/2007_test.txt");
char **paths = (char **)list_to_array(plist);
int classes = layer.classes;
int objectness = layer.objectness;
int background = layer.background;
int num_boxes = sqrt(get_detection_layer_locations(layer));
int j;
FILE **fps = calloc(classes, sizeof(FILE *));
for(j = 0; j < classes; ++j){
char buff[1024];
snprintf(buff, 1024, "%s%s.txt", base, voc_names[j]);
fps[j] = fopen(buff, "w");
}
box *boxes = calloc(num_boxes*num_boxes, sizeof(box));
float **probs = calloc(num_boxes*num_boxes, sizeof(float *));
for(j = 0; j < num_boxes*num_boxes; ++j) probs[j] = calloc(classes, sizeof(float *));
int m = plist->size;
int i=0;
int t;
float thresh = .001;
int nms = 1;
float iou_thresh = .5;
int nthreads = 8;
image *val = calloc(nthreads, sizeof(image));
image *val_resized = calloc(nthreads, sizeof(image));
image *buf = calloc(nthreads, sizeof(image));
image *buf_resized = calloc(nthreads, sizeof(image));
pthread_t *thr = calloc(nthreads, sizeof(pthread_t));
load_args args = {0};
args.w = net.w;
args.h = net.h;
args.type = IMAGE_DATA;
for(t = 0; t < nthreads; ++t){
args.path = paths[i+t];
args.im = &buf[t];
args.resized = &buf_resized[t];
thr[t] = load_data_in_thread(args);
}
time_t start = time(0);
for(i = nthreads; i < m+nthreads; i += nthreads){
fprintf(stderr, "%d\n", i);
for(t = 0; t < nthreads && i+t-nthreads < m; ++t){
pthread_join(thr[t], 0);
val[t] = buf[t];
val_resized[t] = buf_resized[t];
}
for(t = 0; t < nthreads && i+t < m; ++t){
args.path = paths[i+t];
args.im = &buf[t];
args.resized = &buf_resized[t];
thr[t] = load_data_in_thread(args);
}
for(t = 0; t < nthreads && i+t-nthreads < m; ++t){
char *path = paths[i+t-nthreads];
char *id = basecfg(path);
float *X = val_resized[t].data;
float *predictions = network_predict(net, X);
int w = val[t].w;
int h = val[t].h;
convert_yoloplus_detections(predictions, classes, objectness, background, num_boxes, w, h, thresh, probs, boxes);
if (nms) do_nms(boxes, probs, num_boxes*num_boxes, classes, iou_thresh);
print_yoloplus_detections(fps, id, boxes, probs, num_boxes, classes, w, h);
free(id);
free_image(val[t]);
free_image(val_resized[t]);
}
}
fprintf(stderr, "Total Detection Time: %f Seconds\n", (double)(time(0) - start));
}
void test_yoloplus(char *cfgfile, char *weightfile, char *filename, float thresh)
{
network net = parse_network_cfg(cfgfile);
if(weightfile){
load_weights(&net, weightfile);
}
detection_layer layer = get_network_detection_layer(net);
set_batch_network(&net, 1);
srand(2222222);
clock_t time;
char input[256];
while(1){
if(filename){
strncpy(input, filename, 256);
} else {
printf("Enter Image Path: ");
fflush(stdout);
fgets(input, 256, stdin);
strtok(input, "\n");
}
image im = load_image_color(input,0,0);
image sized = resize_image(im, net.w, net.h);
float *X = sized.data;
time=clock();
float *predictions = network_predict(net, X);
printf("%s: Predicted in %f seconds.\n", input, sec(clock()-time));
draw_yoloplus(im, predictions, 7, layer.objectness, "predictions", thresh);
free_image(im);
free_image(sized);
#ifdef OPENCV
cvWaitKey(0);
cvDestroyAllWindows();
#endif
if (filename) break;
}
}
void run_yoloplus(int argc, char **argv)
{
float thresh = find_float_arg(argc, argv, "-thresh", .2);
if(argc < 4){
fprintf(stderr, "usage: %s %s [train/test/valid] [cfg] [weights (optional)]\n", argv[0], argv[1]);
return;
}
char *cfg = argv[3];
char *weights = (argc > 4) ? argv[4] : 0;
char *filename = (argc > 5) ? argv[5]: 0;
if(0==strcmp(argv[2], "test")) test_yoloplus(cfg, weights, filename, thresh);
else if(0==strcmp(argv[2], "train")) train_yoloplus(cfg, weights);
else if(0==strcmp(argv[2], "valid")) validate_yoloplus(cfg, weights);
}