jilstingray / MTWI-2018

The Python implementation of CTPN targeting at MTWI 2018 Challenge 2.

README

The Python implementation of Connectionist Text Proposal Network (CTPN) targeting at MTWI 2018 Challenge 2.

Competition Details

https://tianchi.aliyun.com/competition/entrance/231685/introduction

Environment

• Python 3.x with PyTorch (0.4 or newer)

• cython, configphaser, lmdb, matplotlib, numpy, opencv-python are required.

Run before training

• Run setup_cython.py FIRST:

cd lib
python setup_cython.py build_ext --inplace

• Modify config.py and use it to generate config file for your environment.

• If you've downloaded the original MTWI_2018 dataset from Aliyun, try to use dataset_handler.reorganize_dataset() to reorganize it, since some pictures may not be read as RGB channels.

Predict an image

After training, open predict.py and set MODEL as your trained model's path. Then run predict.py:

python predict.py [path] [mode] [running_mode]

Reference

https://arxiv.org/abs/1609.03605

https://github.com/AstarLight/Lets_OCR/tree/master/detector/ctpn