Fixes for training Yolo on small objects

src/data.c

@@ -292,7 +292,7 @@ void fill_truth_region(char *path, float *truth, int classes, int num_boxes, int
     free(boxes);
 }
 
-void fill_truth_detection(char *path, int num_boxes, float *truth, int classes, int flip, float dx, float dy, float sx, float sy)
+void fill_truth_detection(char *path, int num_boxes, float *truth, int classes, int flip, float dx, float dy, float sx, float sy, int small_object)
 {
     char labelpath[4096];
     find_replace(path, "images", "labels", labelpath);
@@ -305,6 +305,12 @@ void fill_truth_detection(char *path, int num_boxes, float *truth, int classes,
     find_replace(labelpath, ".JPEG", ".txt", labelpath);
     int count = 0;
     box_label *boxes = read_boxes(labelpath, &count);
+	if (small_object == 1) {
+		for (int i = 0; i < count; ++i) {
+			if (boxes[i].w < 0.01) boxes[i].w = 0.01;
+			if (boxes[i].h < 0.01) boxes[i].h = 0.01;
+		}
+	}
     randomize_boxes(boxes, count);
     correct_boxes(boxes, count, dx, dy, sx, sy, flip);
     if(count > num_boxes) count = num_boxes;
@@ -319,7 +325,8 @@ void fill_truth_detection(char *path, int num_boxes, float *truth, int classes,
         h =  boxes[i].h;
         id = boxes[i].id;
 
-        if ((w < .01 || h < .01)) continue;
+		// not detect small objects
+		if ((w < 0.01 || h < 0.01)) continue;
 
         truth[i*5+0] = x;
         truth[i*5+1] = y;
@@ -661,7 +668,7 @@ data load_data_swag(char **paths, int n, int classes, float jitter)
     return d;
 }
 
-data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, int classes, float jitter, float hue, float saturation, float exposure)
+data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, int classes, float jitter, float hue, float saturation, float exposure, int small_object)
 {
     char **random_paths = get_random_paths(paths, n, m);
     int i;
@@ -704,7 +711,7 @@ data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, in
         random_distort_image(sized, hue, saturation, exposure);
         d.X.vals[i] = sized.data;
 
-        fill_truth_detection(random_paths[i], boxes, d.y.vals[i], classes, flip, dx, dy, 1./sx, 1./sy);
+        fill_truth_detection(random_paths[i], boxes, d.y.vals[i], classes, flip, dx, dy, 1./sx, 1./sy, small_object);
 
         free_image(orig);
         free_image(cropped);
@@ -734,7 +741,7 @@ void *load_thread(void *ptr)
     } else if (a.type == REGION_DATA){
         *a.d = load_data_region(a.n, a.paths, a.m, a.w, a.h, a.num_boxes, a.classes, a.jitter, a.hue, a.saturation, a.exposure);
     } else if (a.type == DETECTION_DATA){
-        *a.d = load_data_detection(a.n, a.paths, a.m, a.w, a.h, a.num_boxes, a.classes, a.jitter, a.hue, a.saturation, a.exposure);
+        *a.d = load_data_detection(a.n, a.paths, a.m, a.w, a.h, a.num_boxes, a.classes, a.jitter, a.hue, a.saturation, a.exposure, a.small_object);
     } else if (a.type == SWAG_DATA){
         *a.d = load_data_swag(a.paths, a.n, a.classes, a.jitter);
     } else if (a.type == COMPARE_DATA){

src/data.h

@@ -53,6 +53,7 @@ typedef struct load_args{
     int classes;
     int background;
     int scale;
+	int small_object;
     float jitter;
     float angle;
     float aspect;

src/detector.c

@@ -82,6 +82,7 @@ void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, i
     args.classes = classes;
     args.jitter = jitter;
     args.num_boxes = l.max_boxes;
+	args.small_object = l.small_object;
     args.d = &buffer;
     args.type = DETECTION_DATA;
 	args.threads = 4;// 8;

src/layer.h

@@ -63,6 +63,7 @@ struct layer{
     int out_h, out_w, out_c;
     int n;
     int max_boxes;
+	int small_object;
     int groups;
     int size;
     int side;

src/parser.c

@@ -245,6 +245,7 @@ layer parse_region(list *options, size_params params)
     l.log = option_find_int_quiet(options, "log", 0);
     l.sqrt = option_find_int_quiet(options, "sqrt", 0);
 
+	l.small_object = option_find_int(options, "small_object", 0);
     l.softmax = option_find_int(options, "softmax", 0);
     l.max_boxes = option_find_int_quiet(options, "max",30);
     l.jitter = option_find_float(options, "jitter", .2);