From 2ce6460c79e06caa33eab3991ee3e7fd9f0909d6 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Tue, 10 Apr 2018 18:08:24 +0000
Subject: [PATCH] Remove truth only if smaller than 1x1 pix during training

---
 src/data.c |  177 +++++++++++++++++++++++++++++++++++++++++++++--------------
 1 files changed, 135 insertions(+), 42 deletions(-)

diff --git a/src/data.c b/src/data.c
index 19dca8a..abdb043 100644
--- a/src/data.c
+++ b/src/data.c
@@ -29,7 +29,7 @@
     int i;
     pthread_mutex_lock(&mutex);
     for(i = 0; i < n; ++i){
-        int index = rand()%m;
+        int index = random_gen()%m;
         indexes[i] = index;
         random_paths[i] = paths[index];
         if(i == 0) printf("%s\n", paths[index]);
@@ -45,11 +45,11 @@
     int i;
     pthread_mutex_lock(&mutex);
 	//printf("n = %d \n", n);
-    for(i = 0; i < n; ++i){
-        int index = (rand()*rand())%m;
+    for(i = 0; i < n; ++i){		
+        int index = random_gen() % m;
         random_paths[i] = paths[index];
         //if(i == 0) printf("%s\n", paths[index]);
-		//printf("%s\n", paths[index]);
+		//printf("grp: %s\n", paths[index]);
     }
     pthread_mutex_unlock(&mutex);
     return random_paths;
@@ -115,7 +115,7 @@
     for(i = 0; i < n; ++i){
         image im = load_image_color(paths[i], 0, 0);
         image crop = random_augment_image(im, angle, aspect, min, max, size);
-        int flip = rand()%2;
+        int flip = random_gen()%2;
         if (flip) flip_image(crop);
         random_distort_image(crop, hue, saturation, exposure);
 
@@ -163,7 +163,7 @@
     int i;
     for(i = 0; i < n; ++i){
         box_label swap = b[i];
-        int index = rand()%n;
+        int index = random_gen()%n;
         b[i] = b[index];
         b[index] = swap;
     }
@@ -269,7 +269,7 @@
         h =  boxes[i].h;
         id = boxes[i].id;
 
-        if (w < .01 || h < .01) continue;
+        if (w < .001 || h < .001) continue;
 
         int col = (int)(x*num_boxes);
         int row = (int)(y*num_boxes);
@@ -292,7 +292,8 @@
     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, int net_w, int net_h)
 {
     char labelpath[4096];
     find_replace(path, "images", "labels", labelpath);
@@ -304,13 +305,21 @@
     find_replace(labelpath, ".JPG", ".txt", labelpath);
     find_replace(labelpath, ".JPEG", ".txt", labelpath);
     int count = 0;
+	int i;
     box_label *boxes = read_boxes(labelpath, &count);
+	float lowest_w = 1.F / net_w;
+	float lowest_h = 1.F / net_h;
+	if (small_object == 1) {
+		for (i = 0; i < count; ++i) {
+			if (boxes[i].w < lowest_w) boxes[i].w = lowest_w;
+			if (boxes[i].h < lowest_h) boxes[i].h = lowest_h;
+		}
+	}
     randomize_boxes(boxes, count);
     correct_boxes(boxes, count, dx, dy, sx, sy, flip);
     if(count > num_boxes) count = num_boxes;
     float x,y,w,h;
     int id;
-    int i;
 
     for (i = 0; i < count; ++i) {
         x =  boxes[i].x;
@@ -319,7 +328,10 @@
         h =  boxes[i].h;
         id = boxes[i].id;
 
-        if ((w < .01 || h < .01)) continue;
+		// not detect small objects
+		//if ((w < 0.001F || h < 0.001F)) continue;
+		// if truth (box for object) is smaller than 1x1 pix
+		if ((w < lowest_w || h < lowest_h)) continue;
 
         truth[i*5+0] = x;
         truth[i*5+1] = y;
@@ -524,7 +536,7 @@
         float sx = (float)swidth  / ow;
         float sy = (float)sheight / oh;
 
-        int flip = rand()%2;
+        int flip = random_gen()%2;
         image cropped = crop_image(orig, pleft, ptop, swidth, sheight);
 
         float dx = ((float)pleft/ow)/sx;
@@ -610,7 +622,7 @@
 
 data load_data_swag(char **paths, int n, int classes, float jitter)
 {
-    int index = rand()%n;
+    int index = random_gen()%n;
     char *random_path = paths[index];
 
     image orig = load_image_color(random_path, 0, 0);
@@ -643,7 +655,7 @@
     float sx = (float)swidth  / w;
     float sy = (float)sheight / h;
 
-    int flip = rand()%2;
+    int flip = random_gen()%2;
     image cropped = crop_image(orig, pleft, ptop, swidth, sheight);
 
     float dx = ((float)pleft/w)/sx;
@@ -661,7 +673,18 @@
     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)
+#ifdef OPENCV
+#include "opencv2/highgui/highgui_c.h"
+#include "opencv2/imgproc/imgproc_c.h"
+#include "opencv2/core/version.hpp"
+#ifndef CV_VERSION_EPOCH
+#include "opencv2/videoio/videoio_c.h"
+#include "opencv2/imgcodecs/imgcodecs_c.h"
+#endif
+
+#include "http_stream.h"
+
+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;
@@ -674,18 +697,30 @@
 
     d.y = make_matrix(n, 5*boxes);
     for(i = 0; i < n; ++i){
-        image orig = load_image_color(random_paths[i], 0, 0);
+		const char *filename = random_paths[i];
 
-        int oh = orig.h;
-        int ow = orig.w;
+		int flag = 1;
+		IplImage *src;
+		if ((src = cvLoadImage(filename, flag)) == 0)
+		{
+			fprintf(stderr, "Cannot load image \"%s\"\n", filename);
+			char buff[256];
+			sprintf(buff, "echo %s >> bad.list", filename);
+			system(buff);
+			continue;
+			//exit(0);
+		}
+
+		int oh = src->height;
+		int ow = src->width;
 
         int dw = (ow*jitter);
         int dh = (oh*jitter);
 
-        int pleft  = rand_uniform(-dw, dw);
-        int pright = rand_uniform(-dw, dw);
-        int ptop   = rand_uniform(-dh, dh);
-        int pbot   = rand_uniform(-dh, dh);
+        int pleft  = rand_uniform_strong(-dw, dw);
+        int pright = rand_uniform_strong(-dw, dw);
+        int ptop   = rand_uniform_strong(-dh, dh);
+        int pbot   = rand_uniform_strong(-dh, dh);
 
         int swidth =  ow - pleft - pright;
         int sheight = oh - ptop - pbot;
@@ -693,31 +728,86 @@
         float sx = (float)swidth  / ow;
         float sy = (float)sheight / oh;
 
-        int flip = rand()%2;
-        image cropped = crop_image(orig, pleft, ptop, swidth, sheight);
+        int flip = random_gen()%2;
 
         float dx = ((float)pleft/ow)/sx;
         float dy = ((float)ptop /oh)/sy;
 
-        image sized = resize_image(cropped, w, h);
-        if(flip) flip_image(sized);
-        random_distort_image(sized, hue, saturation, exposure);
-        d.X.vals[i] = sized.data;
+		float dhue = rand_uniform_strong(-hue, hue);
+		float dsat = rand_scale(saturation);
+		float dexp = rand_scale(exposure);
 
-        fill_truth_detection(random_paths[i], boxes, d.y.vals[i], classes, flip, dx, dy, 1./sx, 1./sy);
+		image ai = image_data_augmentation(src, w, h, pleft, ptop, swidth, sheight, flip, jitter, dhue, dsat, dexp);
+		d.X.vals[i] = ai.data;
+		
+		//show_image(ai, "aug");
+		//cvWaitKey(0);
 
-        free_image(orig);
-        free_image(cropped);
+        fill_truth_detection(filename, boxes, d.y.vals[i], classes, flip, dx, dy, 1./sx, 1./sy, small_object, w, h);
+
+		cvReleaseImage(&src);
     }
     free(random_paths);
     return d;
 }
+#else	// OPENCV
+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;
+	data d = { 0 };
+	d.shallow = 0;
 
+	d.X.rows = n;
+	d.X.vals = calloc(d.X.rows, sizeof(float*));
+	d.X.cols = h*w * 3;
+
+	d.y = make_matrix(n, 5 * boxes);
+	for (i = 0; i < n; ++i) {
+		image orig = load_image_color(random_paths[i], 0, 0);
+
+		int oh = orig.h;
+		int ow = orig.w;
+
+		int dw = (ow*jitter);
+		int dh = (oh*jitter);
+
+		int pleft = rand_uniform_strong(-dw, dw);
+		int pright = rand_uniform_strong(-dw, dw);
+		int ptop = rand_uniform_strong(-dh, dh);
+		int pbot = rand_uniform_strong(-dh, dh);
+
+		int swidth = ow - pleft - pright;
+		int sheight = oh - ptop - pbot;
+
+		float sx = (float)swidth / ow;
+		float sy = (float)sheight / oh;
+
+		int flip = random_gen() % 2;
+		image cropped = crop_image(orig, pleft, ptop, swidth, sheight);
+
+		float dx = ((float)pleft / ow) / sx;
+		float dy = ((float)ptop / oh) / sy;
+
+		image sized = resize_image(cropped, w, h);
+		if (flip) flip_image(sized);
+		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, small_object, w, h);
+
+		free_image(orig);
+		free_image(cropped);
+	}
+	free(random_paths);
+	return d;
+}
+#endif	// OPENCV
 
 void *load_thread(void *ptr)
 {
-	srand(time(0));
-    //printf("Loading data: %d\n", rand());
+	//srand(time(0));
+    //printf("Loading data: %d\n", random_gen());
     load_args a = *(struct load_args*)ptr;
     if(a.exposure == 0) a.exposure = 1;
     if(a.saturation == 0) a.saturation = 1;
@@ -734,7 +824,7 @@
     } 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){
@@ -742,6 +832,9 @@
     } else if (a.type == IMAGE_DATA){
         *(a.im) = load_image_color(a.path, 0, 0);
         *(a.resized) = resize_image(*(a.im), a.w, a.h);
+	}else if (a.type == LETTERBOX_DATA) {
+		*(a.im) = load_image_color(a.path, 0, 0);
+		*(a.resized) = letterbox_image(*(a.im), a.w, a.h);
     } else if (a.type == TAG_DATA){
         *a.d = load_data_tag(a.paths, a.n, a.m, a.classes, a.min, a.max, a.size, a.angle, a.aspect, a.hue, a.saturation, a.exposure);
     }
@@ -760,7 +853,7 @@
 
 void *load_threads(void *ptr)
 {
-	srand(time(0));
+	//srand(time(0));
     int i;
     load_args args = *(load_args *)ptr;
     if (args.threads == 0) args.threads = 1;
@@ -855,7 +948,7 @@
     for(i = 0; i < n; ++i){
         image im = load_image_color(paths[i], 0, 0);
         image crop = random_crop_image(im, w*scale, h*scale);
-        int flip = rand()%2;
+        int flip = random_gen()%2;
         if (flip) flip_image(crop);
         image resize = resize_image(crop, w, h);
         d.X.vals[i] = resize.data;
@@ -960,8 +1053,8 @@
     for(i = 0; i < 10000; ++i){
         unsigned char bytes[3073];
         fread(bytes, 1, 3073, fp);
-        int class = bytes[0];
-        y.vals[i][class] = 1;
+        int class_id = bytes[0];
+        y.vals[i][class_id] = 1;
         for(j = 0; j < X.cols; ++j){
             X.vals[i][j] = (double)bytes[j+1];
         }
@@ -977,7 +1070,7 @@
 {
     int j;
     for(j = 0; j < n; ++j){
-        int index = rand()%d.X.rows;
+        int index = random_gen()%d.X.rows;
         memcpy(X+j*d.X.cols, d.X.vals[index], d.X.cols*sizeof(float));
         memcpy(y+j*d.y.cols, d.y.vals[index], d.y.cols*sizeof(float));
     }
@@ -1024,8 +1117,8 @@
         for(i = 0; i < 10000; ++i){
             unsigned char bytes[3073];
             fread(bytes, 1, 3073, fp);
-            int class = bytes[0];
-            y.vals[i+b*10000][class] = 1;
+            int class_id = bytes[0];
+            y.vals[i+b*10000][class_id] = 1;
             for(j = 0; j < X.cols; ++j){
                 X.vals[i+b*10000][j] = (double)bytes[j+1];
             }
@@ -1090,7 +1183,7 @@
 {
     int i;
     for(i = d.X.rows-1; i > 0; --i){
-        int index = rand()%i;
+        int index = random_gen()%i;
         float *swap = d.X.vals[index];
         d.X.vals[index] = d.X.vals[i];
         d.X.vals[i] = swap;
@@ -1154,7 +1247,7 @@
 
     int i;
     for(i = 0; i < num; ++i){
-        int index = rand()%d.X.rows;
+        int index = random_gen()%d.X.rows;
         r.X.vals[i] = d.X.vals[index];
         r.y.vals[i] = d.y.vals[index];
     }

--
Gitblit v1.10.0