Lineae ex punctis

Nov 17, 2018       tags: |

Note: Originally I used the following method which involves running the YOLOv3 model natively on darknet and requires some modifications to be made in order to output image paths and object coordinates in a way that can be used by my Python script. You can save yourself a lot of time if you run YOLOv3 directly with OpenCV from Python. To do that please return to the main blog entry

Car detection — old method

The photographs were batch-cropped (mogrify -crop 4000x2050+0+750 *.jpg) and processed by YOLOv3 (with COCO weights). If you want to replicate this project, you can follow the link for detailed YOLOv3 instructions or simply run the following commands (assuming you keep git repositories in ~/git):

cd ~/git
git clone https://github.com/pjreddie/darknet
cd darknet
wget https://pjreddie.com/media/files/yolov3.weights

1 Edit void draw_detections(...) in src/image.c:

1.1 At the top, after int i,j; add

int max_prob_class;
float max_prob = 0;

1.2 Insert

if (dets[i].prob[j] > max_prob) {
	max_prob = dets[i].prob[j];
	max_prob_class = j; 
}

in the following cycle

for(j = 0; j < classes; ++j){
	if (dets[i].prob[j] > thresh){
		if (class < 0) {
			strcat(labelstr, names[j]);
			class = j;
		} else {
			strcat(labelstr, ", ");
			strcat(labelstr, names[j]);
		}
		// <<< insert HERE >>>   
	}
}

1.3 Add

printf("{\"object\": \"%s\", \"probability\": \"%.2f\", \"x0\": \"%d\", \"y0\": \"%d\", \"x1\": \"%d\", \"y1\": \"%d\"}\n", names[max_prob_class], max_prob, left, top, right, bot); 

below

if(left < 0) left = 0;
if(right > im.w-1) right = im.w-1;
if(top < 0) top = 0;
if(bot > im.h-1) bot = im.h-1;

2 Edit void test_detector(...) in examples/detector.c: replace printf("%s: Predicted in %f seconds.\n", input, what_time_is_it_now()-time); with printf("{\"path\": \"%s\"}\n", input);

Run make after saving the changes to compile darknet.

In order to run YOLOv3 with multiple images, we need a file containing the images’ paths. It can be created like this:

cd <image directory>
ls -d -1 $PWD/*.* | grep '.JPG' > imagelist.txt

Finally, image recognition can be run:

cd <darknet directory>
./darknet detect cfg/yolov3.cfg yolov3.weights < /~/Desktop/images/imagelist.txt

The terminal output needs to be saved into detected_objects.txt in the directory of choice.