python, pytorch, captcha recognize, cnn This is learning project to understand how machine learning works. Aim - create and tech network to do text on picture recognition.

Whole process split by few steps:

1. We should grab/download some images to use them as train/validate set, please use grabber.py as an entry point, inside file you can tweak how much data do you need
2. Your downloaded pictures should be described in the desc.json where you put picture name and exact text it contains
3. Run image_transform.py to do image normalization, it will remove picture noise, convert to black/white and crop to appropriate size
4. Run neuro_image.py to actually start teaching your network (pay attention, currently network storing commented out)
5. Run model.py to load pre-trained network and try to recognize real picture

Didn't try to make it as a finally solution, so all files probably requires minimal tweaks.