Solve Image based captchas for OwObot
train.py -> To actually train the model. No pre-processing of images was done (was lazy). Perhaps you should consider preprocessing, as that may fetch better results.
test_onnx.p -> After exporting to an ONNX model, we can test the captcha solver with this file.
examples folder contains 5 example captchas from OwO-Bot
All captcha images in use in my dataset was fetched with the help of https://github.com/Tyrrrz/DiscordChatExporter https://github.com/wkentaro/labelme was used to annotate captcha images.
Took around 2~ days to get this to 90~ % accuracy with only 500~ images used for training. With better control of images in use, perhaps it could be reduced to 300 as I managed to achieve80% accuracy with just 300.
Feel free to use this mode in compliance with GNU GPL V3 licence :>