From 8f1f5cbf8321b6b313d8f455d596290e7b8bb3f7 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Sat, 24 Mar 2018 20:35:05 +0000
Subject: [PATCH] 3.5x times accelerated data augmentation using OpenCV for training Yolo on DGX and multi-GPU
---
src/data.c | 100 ++++++++++++++++++++++++++++++++++++++++++++-----
1 files changed, 89 insertions(+), 11 deletions(-)
diff --git a/src/data.c b/src/data.c
index 379756a..d9cedf5 100644
--- a/src/data.c
+++ b/src/data.c
@@ -668,6 +668,17 @@
return d;
}
+#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);
@@ -681,10 +692,22 @@
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);
@@ -701,25 +724,80 @@
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;
+ 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, small_object);
+ 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);
+
+ 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);
+
+ free_image(orig);
+ free_image(cropped);
+ }
+ free(random_paths);
+ return d;
+}
+#endif // OPENCV
void *load_thread(void *ptr)
{
--
Gitblit v1.10.0