hi

Makefile

@@ -1,4 +1,4 @@
-GPU=0
+GPU=1
 OPENCV=1
 CUDNN=0
 DEBUG=0
@@ -41,7 +41,7 @@ CFLAGS+= -DCUDNN
 LDFLAGS+= -lcudnn
 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 layer.o compare.o classifier.o local_layer.o swag.o shortcut_layer.o activation_layer.o rnn_layer.o gru_layer.o rnn.o rnn_vid.o crnn_layer.o coco_demo.o tag.o cifar.o yolo_demo.o go.o batchnorm_layer.o art.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 layer.o compare.o classifier.o local_layer.o swag.o shortcut_layer.o activation_layer.o rnn_layer.o gru_layer.o rnn.o rnn_vid.o crnn_layer.o demo.o tag.o cifar.o go.o batchnorm_layer.o art.o
 ifeq ($(GPU), 1) 
 LDFLAGS+= -lstdc++ 
 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

cfg/yolo-coco.cfg

@@ -1,6 +1,6 @@
 [net]
-batch=64
-subdivisions=4
+batch=1
+subdivisions=1
 height=448
 width=448
 channels=3

cfg/yolo.cfg

@@ -1,6 +1,6 @@
 [net]
-batch=64
-subdivisions=64
+batch=1
+subdivisions=1
 height=448
 width=448
 channels=3
@@ -13,14 +13,6 @@ steps=200,400,600,20000,30000
 scales=2.5,2,2,.1,.1
 max_batches = 40000
 
-[crop]
-crop_width=448
-crop_height=448
-flip=0
-angle=0
-saturation = 1.5
-exposure = 1.5
-
 [convolutional]
 filters=64
 size=7
@@ -211,9 +203,6 @@ activation=leaky
 output=4096
 activation=leaky
 
-[dropout]
-probability=.5
-
 [connected]
 output= 1470
 activation=linear

src/activation_kernels.cu

@@ -8,6 +8,18 @@ extern "C" {
 }
 
 
+__device__ float lhtan_activate_kernel(float x)
+{
+    if(x < 0) return .001*x;
+    if(x > 1) return .001*(x-1) + 1;
+    return x;
+}
+__device__ float lhtan_gradient_kernel(float x)
+{
+    if(x > 0 && x < 1) return 1;
+    return .001;
+}
+
 __device__ float hardtan_activate_kernel(float x)
 {
     if (x < -1) return -1;
@@ -89,6 +101,8 @@ __device__ float activate_kernel(float x, ACTIVATION a)
             return stair_activate_kernel(x);
         case HARDTAN:
             return hardtan_activate_kernel(x);
+        case LHTAN:
+            return lhtan_activate_kernel(x);
     }
     return 0;
 }
@@ -120,6 +134,8 @@ __device__ float gradient_kernel(float x, ACTIVATION a)
             return stair_gradient_kernel(x);
         case HARDTAN:
             return hardtan_gradient_kernel(x);
+        case LHTAN:
+            return lhtan_gradient_kernel(x);
     }
     return 0;
 }

src/activations.c

@@ -32,6 +32,8 @@ char *get_activation_string(ACTIVATION a)
             return "stair";
         case HARDTAN:
             return "hardtan";
+        case LHTAN:
+            return "lhtan";
         default:
             break;
     }
@@ -47,6 +49,7 @@ ACTIVATION get_activation(char *s)
     if (strcmp(s, "relie")==0) return RELIE;
     if (strcmp(s, "plse")==0) return PLSE;
     if (strcmp(s, "hardtan")==0) return HARDTAN;
+    if (strcmp(s, "lhtan")==0) return LHTAN;
     if (strcmp(s, "linear")==0) return LINEAR;
     if (strcmp(s, "ramp")==0) return RAMP;
     if (strcmp(s, "leaky")==0) return LEAKY;
@@ -83,6 +86,8 @@ float activate(float x, ACTIVATION a)
             return stair_activate(x);
         case HARDTAN:
             return hardtan_activate(x);
+        case LHTAN:
+            return lhtan_activate(x);
     }
     return 0;
 }
@@ -122,6 +127,8 @@ float gradient(float x, ACTIVATION a)
             return stair_gradient(x);
         case HARDTAN:
             return hardtan_gradient(x);
+        case LHTAN:
+            return lhtan_gradient(x);
     }
     return 0;
 }

src/activations.h

@@ -4,7 +4,7 @@
 #include "math.h"
 
 typedef enum{
-    LOGISTIC, RELU, RELIE, LINEAR, RAMP, TANH, PLSE, LEAKY, ELU, LOGGY, STAIR, HARDTAN
+    LOGISTIC, RELU, RELIE, LINEAR, RAMP, TANH, PLSE, LEAKY, ELU, LOGGY, STAIR, HARDTAN, LHTAN
 }ACTIVATION;
 
 ACTIVATION get_activation(char *s);
@@ -47,6 +47,18 @@ static inline float plse_activate(float x)
     return .125*x + .5;
 }
 
+static inline float lhtan_activate(float x)
+{
+    if(x < 0) return .001*x;
+    if(x > 1) return .001*(x-1) + 1;
+    return x;
+}
+static inline float lhtan_gradient(float x)
+{
+    if(x > 0 && x < 1) return 1;
+    return .001;
+}
+
 static inline float hardtan_gradient(float x)
 {
     if (x > -1 && x < 1) return 1;

src/coco.c

@@ -6,11 +6,14 @@
 #include "utils.h"
 #include "parser.h"
 #include "box.h"
+#include "demo.h"
 
 #ifdef OPENCV
 #include "opencv2/highgui/highgui_c.h"
 #endif
 
+void convert_detections(float *predictions, int classes, int num, int square, int side, int w, int h, float thresh, float **probs, box *boxes, int only_objectness);
+
 char *coco_classes[] = {"person","bicycle","car","motorcycle","airplane","bus","train","truck","boat","traffic light","fire hydrant","stop sign","parking meter","bench","bird","cat","dog","horse","sheep","cow","elephant","bear","zebra","giraffe","backpack","umbrella","handbag","tie","suitcase","frisbee","skis","snowboard","sports ball","kite","baseball bat","baseball glove","skateboard","surfboard","tennis racket","bottle","wine glass","cup","fork","knife","spoon","bowl","banana","apple","sandwich","orange","broccoli","carrot","hot dog","pizza","donut","cake","chair","couch","potted plant","bed","dining table","toilet","tv","laptop","mouse","remote","keyboard","cell phone","microwave","oven","toaster","sink","refrigerator","book","clock","vase","scissors","teddy bear","hair drier","toothbrush"};
 
 int coco_ids[] = {1,2,3,4,5,6,7,8,9,10,11,13,14,15,16,17,18,19,20,21,22,23,24,25,27,28,31,32,33,34,35,36,37,38,39,40,41,42,43,44,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,67,70,72,73,74,75,76,77,78,79,80,81,82,84,85,86,87,88,89,90};
@@ -98,34 +101,6 @@ void train_coco(char *cfgfile, char *weightfile)
     save_weights(net, buff);
 }
 
-void convert_coco_detections(float *predictions, int classes, int num, int square, int side, int w, int h, float thresh, float **probs, box *boxes, int only_objectness)
-{
-    int i,j,n;
-    //int per_cell = 5*num+classes;
-    for (i = 0; i < side*side; ++i){
-        int row = i / side;
-        int col = i % side;
-        for(n = 0; n < num; ++n){
-            int index = i*num + n;
-            int p_index = side*side*classes + i*num + n;
-            float scale = predictions[p_index];
-            int box_index = side*side*(classes + num) + (i*num + n)*4;
-            boxes[index].x = (predictions[box_index + 0] + col) / side * w;
-            boxes[index].y = (predictions[box_index + 1] + row) / side * h;
-            boxes[index].w = pow(predictions[box_index + 2], (square?2:1)) * w;
-            boxes[index].h = pow(predictions[box_index + 3], (square?2:1)) * h;
-            for(j = 0; j < classes; ++j){
-                int class_index = i*classes;
-                float prob = scale*predictions[class_index+j];
-                probs[index][j] = (prob > thresh) ? prob : 0;
-            }
-            if(only_objectness){
-                probs[index][0] = scale;
-            }
-        }
-    }
-}
-
 void print_cocos(FILE *fp, int image_id, box *boxes, float **probs, int num_boxes, int classes, int w, int h)
 {
     int i, j;
@@ -235,7 +210,7 @@ void validate_coco(char *cfgfile, char *weightfile)
             float *predictions = network_predict(net, X);
             int w = val[t].w;
             int h = val[t].h;
-            convert_coco_detections(predictions, classes, l.n, square, side, w, h, thresh, probs, boxes, 0);
+            convert_detections(predictions, classes, l.n, square, side, w, h, thresh, probs, boxes, 0);
             if (nms) do_nms_sort(boxes, probs, side*side*l.n, classes, iou_thresh);
             print_cocos(fp, image_id, boxes, probs, side*side*l.n, classes, w, h);
             free_image(val[t]);
@@ -298,7 +273,7 @@ void validate_coco_recall(char *cfgfile, char *weightfile)
         image sized = resize_image(orig, net.w, net.h);
         char *id = basecfg(path);
         float *predictions = network_predict(net, sized.data);
-        convert_coco_detections(predictions, classes, l.n, square, side, 1, 1, thresh, probs, boxes, 1);
+        convert_detections(predictions, classes, l.n, square, side, 1, 1, thresh, probs, boxes, 1);
         if (nms) do_nms(boxes, probs, side*side*l.n, 1, nms_thresh);
 
         char *labelpath = find_replace(path, "images", "labels");
@@ -370,7 +345,7 @@ void test_coco(char *cfgfile, char *weightfile, char *filename, float thresh)
         time=clock();
         float *predictions = network_predict(net, X);
         printf("%s: Predicted in %f seconds.\n", input, sec(clock()-time));
-        convert_coco_detections(predictions, l.classes, l.n, l.sqrt, l.side, 1, 1, thresh, probs, boxes, 0);
+        convert_detections(predictions, l.classes, l.n, l.sqrt, l.side, 1, 1, thresh, probs, boxes, 0);
         if (nms) do_nms_sort(boxes, probs, l.side*l.side*l.n, l.classes, nms);
         draw_detections(im, l.side*l.side*l.n, thresh, boxes, probs, coco_classes, coco_labels, 80);
         show_image(im, "predictions");
@@ -386,16 +361,6 @@ void test_coco(char *cfgfile, char *weightfile, char *filename, float thresh)
     }
 }
 
-void demo_coco(char *cfgfile, char *weightfile, float thresh, int cam_index, char *filename);
-static void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, char* filename)
-{
-    #if defined(OPENCV)
-        demo_coco(cfgfile, weightfile, thresh, cam_index, filename);
-    #else
-        fprintf(stderr, "Need to compile with OpenCV for demo.\n");
-    #endif
-}
-
 void run_coco(int argc, char **argv)
 {
     int i;
@@ -406,7 +371,6 @@ void run_coco(int argc, char **argv)
     }
     float thresh = find_float_arg(argc, argv, "-thresh", .2);
     int cam_index = find_int_arg(argc, argv, "-c", 0);
-    char *file = find_char_arg(argc, argv, "-file", 0);
 
     if(argc < 4){
         fprintf(stderr, "usage: %s %s [train/test/valid] [cfg] [weights (optional)]\n", argv[0], argv[1]);
@@ -420,5 +384,5 @@ void run_coco(int argc, char **argv)
     else if(0==strcmp(argv[2], "train")) train_coco(cfg, weights);
     else if(0==strcmp(argv[2], "valid")) validate_coco(cfg, weights);
     else if(0==strcmp(argv[2], "recall")) validate_coco_recall(cfg, weights);
-    else if(0==strcmp(argv[2], "demo")) demo(cfg, weights, thresh, cam_index, file);
+    else if(0==strcmp(argv[2], "demo")) demo(cfg, weights, thresh, cam_index, filename, coco_classes, coco_labels, 80);
 }

src/convolutional_kernels.cu

@@ -71,8 +71,6 @@ void binarize_filters_gpu(float *filters, int n, int size, float *binary)
 
 void forward_convolutional_layer_gpu(convolutional_layer l, network_state state)
 {
-    int i;
-
     fill_ongpu(l.outputs*l.batch, 0, l.output_gpu, 1);
     if(l.binary){
         binarize_filters_gpu(l.filters_gpu, l.n, l.c*l.size*l.size, l.binary_filters_gpu);
@@ -103,6 +101,7 @@ void forward_convolutional_layer_gpu(convolutional_layer l, network_state state)
                 l.output_gpu);
 
 #else
+    int i;
     int m = l.n;
     int k = l.size*l.size*l.c;
     int n = l.out_w*l.out_h;

src/demo.c

@@ -5,17 +5,22 @@
 #include "parser.h"
 #include "box.h"
 #include "image.h"
+#include "demo.h"
 #include <sys/time.h>
 
-#define FRAMES 1
+#define FRAMES 3
 
 #ifdef OPENCV
 #include "opencv2/highgui/highgui_c.h"
 #include "opencv2/imgproc/imgproc_c.h"
-void convert_coco_detections(float *predictions, int classes, int num, int square, int side, int w, int h, float thresh, float **probs, box *boxes, int only_objectness);
+void convert_detections(float *predictions, int classes, int num, int square, int side, int w, int h, float thresh, float **probs, box *boxes, int only_objectness);
 
-extern char *coco_classes[];
-extern image coco_labels[];
+#define DELAY 0
+static int delay = DELAY;
+
+static char **demo_names;
+static image *demo_labels;
+static int demo_classes;
 
 static float **probs;
 static box *boxes;
@@ -24,7 +29,7 @@ static image in   ;
 static image in_s ;
 static image det  ;
 static image det_s;
-static image disp ;
+static image disp = {0};
 static CvCapture * cap;
 static float fps = 0;
 static float demo_thresh = 0;
@@ -34,14 +39,22 @@ static int demo_index = 0;
 static image images[FRAMES];
 static float *avg;
 
-void *fetch_in_thread_coco(void *ptr)
+void *fetch_in_thread(void *ptr)
 {
     in = get_image_from_stream(cap);
-    in_s = resize_image(in, net.w, net.h);
+    if(!in.data){
+        in = disp;
+        if(delay == DELAY) error("Stream closed.");
+    }else{
+        if(disp.data){
+            free_image(disp);
+        }
+        in_s = resize_image(in, net.w, net.h);
+    }
     return 0;
 }
 
-void *detect_in_thread_coco(void *ptr)
+void *detect_in_thread(void *ptr)
 {
     float nms = .4;
 
@@ -50,28 +63,47 @@ void *detect_in_thread_coco(void *ptr)
     float *prediction = network_predict(net, X);
 
     memcpy(predictions[demo_index], prediction, l.outputs*sizeof(float));
-    mean_arrays(predictions, FRAMES, l.outputs, avg);
+    if(delay == DELAY){
+        mean_arrays(predictions, FRAMES, l.outputs, avg);
+    }
 
     free_image(det_s);
-    convert_coco_detections(avg, l.classes, l.n, l.sqrt, l.side, 1, 1, demo_thresh, probs, boxes, 0);
+    convert_detections(avg, l.classes, l.n, l.sqrt, l.side, 1, 1, demo_thresh, probs, boxes, 0);
     if (nms > 0) do_nms(boxes, probs, l.side*l.side*l.n, l.classes, nms);
     printf("\033[2J");
     printf("\033[1;1H");
-    printf("\nFPS:%.0f\n",fps);
+    printf("\nFPS:%.1f\n",fps);
     printf("Objects:\n\n");
 
     images[demo_index] = det;
     det = images[(demo_index + FRAMES/2 + 1)%FRAMES];
     demo_index = (demo_index + 1)%FRAMES;
 
-    draw_detections(det, l.side*l.side*l.n, demo_thresh, boxes, probs, coco_classes, coco_labels, 80);
+    draw_detections(det, l.side*l.side*l.n, demo_thresh, boxes, probs, demo_names, demo_labels, demo_classes);
+    if(delay == 0){
+        delay = DELAY;
+    } else {
+        --delay;
+    }
     return 0;
 }
 
-void demo_coco(char *cfgfile, char *weightfile, float thresh, int cam_index, const char *filename)
+double get_wall_time()
 {
+    struct timeval time;
+    if (gettimeofday(&time,NULL)){
+        return 0;
+    }
+    return (double)time.tv_sec + (double)time.tv_usec * .000001;
+}
+
+void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const char *filename, char **names, image *labels, int classes)
+{
+    demo_names = names;
+    demo_labels = labels;
+    demo_classes = classes;
     demo_thresh = thresh;
-    printf("COCO demo\n");
+    printf("Demo\n");
     net = parse_network_cfg(cfgfile);
     if(weightfile){
         load_weights(&net, weightfile);
@@ -102,44 +134,46 @@ void demo_coco(char *cfgfile, char *weightfile, float thresh, int cam_index, con
     pthread_t fetch_thread;
     pthread_t detect_thread;
 
-    fetch_in_thread_coco(0);
+    fetch_in_thread(0);
     det = in;
     det_s = in_s;
 
-    fetch_in_thread_coco(0);
-    detect_in_thread_coco(0);
+    fetch_in_thread(0);
+    detect_in_thread(0);
     disp = det;
     det = in;
     det_s = in_s;
 
     for(j = 0; j < FRAMES/2; ++j){
-        fetch_in_thread_coco(0);
-        detect_in_thread_coco(0);
+        fetch_in_thread(0);
+        detect_in_thread(0);
         disp = det;
         det = in;
         det_s = in_s;
     }
 
     int count = 0;
-    cvNamedWindow("COCO", CV_WINDOW_NORMAL); 
-    cvMoveWindow("COCO", 0, 0);
-    cvResizeWindow("COCO", 1352, 1013);
+    cvNamedWindow("Demo", CV_WINDOW_NORMAL); 
+    cvMoveWindow("Demo", 0, 0);
+    cvResizeWindow("Demo", 1352, 1013);
+
+    double before = get_wall_time();
 
     while(1){
         ++count;
-        struct timeval tval_before, tval_after, tval_result;
-        gettimeofday(&tval_before, NULL);
-        if(pthread_create(&fetch_thread, 0, fetch_in_thread_coco, 0)) error("Thread creation failed");
-        if(pthread_create(&detect_thread, 0, detect_in_thread_coco, 0)) error("Thread creation failed");
-        show_image(disp, "COCO");
-        /*
-        char buff[256];
-        sprintf(buff, "/home/pjreddie/coco/coco_%05d", count);
-        save_image(disp, buff);
-        */
+        if(pthread_create(&fetch_thread, 0, fetch_in_thread, 0)) error("Thread creation failed");
+        if(pthread_create(&detect_thread, 0, detect_in_thread, 0)) error("Thread creation failed");
+        //fetch_in_thread(0);
+        //detect_in_thread(0);
 
-        free_image(disp);
-        cvWaitKey(10);
+        show_image(disp, "Demo");
+        cvWaitKey(1);
+        //char buff[256];
+        //sprintf(buff, "coco/coco_%05d", count);
+        //save_image(disp, buff);
+
+        //free_image(disp);
+        //cvWaitKey(10);
         pthread_join(fetch_thread, 0);
         pthread_join(detect_thread, 0);
 
@@ -147,15 +181,18 @@ void demo_coco(char *cfgfile, char *weightfile, float thresh, int cam_index, con
         det   = in;
         det_s = in_s;
 
-        gettimeofday(&tval_after, NULL);
-        timersub(&tval_after, &tval_before, &tval_result);
-        float curr = 1000000.f/((long int)tval_result.tv_usec);
-        fps = .9*fps + .1*curr;
+        if(delay == DELAY){
+            double after = get_wall_time();
+            float curr = 1./(after - before);
+            fps = curr;
+            before = after;
+        }
     }
 }
 #else
-void demo_coco(char *cfgfile, char *weightfile, float thresh, int cam_index){
-    fprintf(stderr, "YOLO-COCO demo needs OpenCV for webcam images.\n");
+void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const char *filename, char **names, image *labels, int classes)
+{
+    fprintf(stderr, "Demo needs OpenCV for webcam images.\n");
 }
 #endif
 

src/demo.h

@@ -0,0 +1,7 @@
+#ifndef DEMO
+#define DEMO
+
+#include "image.h"
+void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const char *filename, char **names, image *labels, int classes);
+
+#endif

src/detection_layer.c

@@ -53,8 +53,6 @@ void forward_detection_layer(const detection_layer l, network_state state)
                 softmax_array(l.output + index + offset, l.classes, 1,
                         l.output + index + offset);
             }
-            int offset = locations*l.classes;
-            activate_array(l.output + index + offset, locations*l.n*(1+l.coords), LOGISTIC);
         }
     }
     if(state.train){
@@ -133,11 +131,9 @@ void forward_detection_layer(const detection_layer l, network_state state)
                         best_index = 0;
                     }
                 }
-                /*
-                if(1 && *(state.net.seen) < 100000){
+                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;
@@ -175,10 +171,6 @@ void forward_detection_layer(const detection_layer l, network_state state)
                 avg_iou += iou;
                 ++count;
             }
-            if(l.softmax){
-                gradient_array(l.output + index + locations*l.classes, locations*l.n*(1+l.coords), 
-                        LOGISTIC, l.delta + index + locations*l.classes);
-            }
         }
 
         if(0){
@@ -208,6 +200,7 @@ void forward_detection_layer(const detection_layer l, network_state state)
         }
 
 
+        *(l.cost) = pow(mag_array(l.delta, l.outputs * l.batch), 2);
 
 
         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);

src/image.c

@@ -365,6 +365,7 @@ void show_image_cv(image p, const char *name)
     image get_image_from_stream(CvCapture *cap)
     {
         IplImage* src = cvQueryFrame(cap);
+        if (!src) return make_empty_image(0,0,0);
         image im = ipl_to_image(src);
         rgbgr_image(im);
         return im;

src/layer.h

@@ -88,6 +88,7 @@ struct layer{
     float object_scale;
     float noobject_scale;
     float class_scale;
+    int random;
 
     int dontload;
     int dontloadscales;

src/network.c

@@ -64,6 +64,7 @@ float get_current_rate(network net)
         case EXP:
             return net.learning_rate * pow(net.gamma, batch_num);
         case POLY:
+            if (batch_num < net.burn_in) return net.learning_rate * pow((float)batch_num / net.burn_in, net.power);
             return net.learning_rate * pow(1 - (float)batch_num / net.max_batches, net.power);
         case RANDOM:
             return net.learning_rate * pow(rand_uniform(0,1), net.power);