Captcha solver using YOLO v3. Has an accuracy of 98.166% on the test dataset.
-
The trained weights and config file can be found here.
-
Save the trained final weights, yolov3 configuration file and captcha names file in a folder named data.
-
The annotation of dataset is done manually using BBox Label Tool. We find the object's centre x and y coordinates and normalized width and height.
Object-d | x-coordinate | y-coordinate | width | height |
---|---|---|---|---|
0 | 0.285185 | 0.528571 | 0.111111 | 0.600000 |
0 | 0.396296 | 0.528571 | 0.111111 | 0.542857 |
0 | 0.500000 | 0.457143 | 0.111111 | 0.571429 |
1 | 0.607407 | 0.514286 | 0.118519 | 0.571429 |
2 | 0.725926 | 0.557143 | 0.118519 | 0.600000 |
python3 captcha-single.py --image test.jpg # replace with your image
python3 captcha.py #for running on images in test/Solved-600 folder