The full code has been released in the Adet, including the models of CTW1500 and Total-text, all the training data we used, evaluation scripts, results of detection, etc. More updates will also be conducted on Adet as well. This repo will not be maintained anymore.

Paper Link. Note this paper is not the final version. We will update soon.

Check Installation for installation instructions.

bash vis_demo.sh

We assume that your symlinked datasets/totaltext directory has the following structure:

totaltext
|_ test_images
|  |_ 0000000.jpg
|  |_ ...
|  |_ 0000299.jpg
|_ annotations
|  |_ total_test.json 

Model [Google Drive]

Totaltext test data [Google Drive]

Syntext-150k (Part1: 54,327 [imgs][annos]. Part2: 94,723 [imgs][annos].)

python tools/tests/single_demo_bezier.py

Bezier-curve generated script link.

CTW1500 visualization results [link] (original training images can be downloaded Here)

Totaltext visualization results [link]

@article{liu2020abcnet,
  title   =  {ABCNet: Real-time Scene Text Spotting with Adaptive Bezier-Curve Network},
  author  =  {Liu, Yuliang* and Chen, Hao* and Shen, Chunhua and He, Tong and Jin, Lianwen and Wang, Liangwei},
  journal =  {arXiv preprint arXiv:2002.10200},
  year    =  {2020}
}

• *represents the authors contributed equally.

Any suggestion is welcome. Please send email to [email protected] or [email protected]

For commercial purpose usage, please contact Dr. Lianwen Jin: [email protected]

Copyright 2019, Deep Learning and Vision Computing Lab, South China China University of Technology. http://www.dlvc-lab.net