verfication code recognization with pytorch 0.4
- run verif_gen.py to generate training set and test set.
- python -m visdom.server and run train.py.
- eval.py to eval the precision.
- test.py to detect image.
the best result is 91.30% on test set.
trained weight:
Google drive: https://drive.google.com/file/d/1rT1k7vKqj6rPbHFRnKKtdBlKsEDXukmO/view?usp=sharing
百度网盘：链接: https://pan.baidu.com/s/1jGQ1tM3YYyxBFpVY7zTCOQ 提取码: d3pe
loss figure:
prec figure:
