A keras cnn example: use cnn crack BT Dashboard(宝塔面板) captcha code.
follow the steps and you will get a nearly 100% acc model of BTDashboard captchacode.
Because of BTDashboard only works on python2, so this project only support python2.7, so you should have python2.7 installed.
clone this repo.
git clone https://github.com/fiht/test_cnnAll packages we need is at requirements.txt , I strongly suggest you use virtualenv manage packages.
you can train the model by you self or just use the model I've trained to predict captchacode.
run run_cnn.py to train the model. As default, it will train 64 epoll, you can change it by edit the source code at cnn.py line 125. each model it would save trained result.feel easy to kill it if it can predict well or cost too much time.
I've trained the model with 1024 epoll. you can use it directly. Run python predict.py to know how to use.
visit my blog get more detail.
