• Stars
    star
    292
  • Rank 142,152 (Top 3 %)
  • Language
    Python
  • Created almost 6 years ago
  • Updated over 5 years ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

This is a pytorch implementation of CTPN(Detecting Text in Natural Image with Connectionist Text Proposal Network). You may want to finetune from: https://drive.google.com/open?id=1JHhI4sEIXfs5gDa1I9AgJBY477HTzAd0

Pytorch CTPN

update 19-03-20 wed: android ocr

This is a pytorch implementation of CTPN(Detecting Text in Natural Image with Connectionist Text Proposal Network).Inspired by keras-ocr.

Training log is available:Training Log(Chinese)

model size
keras-CTPN 142M
pytorch-CTPN 67.6M

train

  • download ctpn model weights (converted from keras ctpn weights) ctpn_keras_weights.pth.tar from dropbox, and move it to ./checkpoints/ (For a number of reasons, the pretrained weights will no longer be available.Thanks for your attention.)
  • download VOC2007_text_detection Chinese Text Detection dataset and move it to ./images/
  • run python ctpn_train.py --image-dir image_dir --labels-dir labels_dir --num-workers num_workers

predict

  • download the pretrained weights from dropbox
  • Please refer to predict.py for more details.

results

Training Log(Chinese)

Android DEMO

These days, I'm working on deploying this model on Android devices.you can check the results from here.

Android text recognition 4-23

Find out that adopting skew transform can significantly improve recognition accuracy.(It may take a few seconds, heavily depends on your harware and input image size)

reco

reference

Licence

MIT License