This code repository contains the implementations of the paper Text Perceptron: Towards End-to-End Arbitrary-Shaped Text Spotting (AAAI 2020 oral).
To date, the detection part in the Text Perceptron is re-implemented based on the common repository Davar-OCR, and the recognization part will be released in the future.
The model's implementation is in mmdetection/third_party/text_perceptron. The running scritps/demo is in demo/text_perceptron_det/.
Before running the demo code, you should compile and install corresponding dependencies in mmdetection/third_party/.
Step 1: Install Davar-OCR (complie mmdetection and mmcv)
>>> cd $OCR_ROOT$
>>> bash setup.sh Step 2: Uncomment corresponding importing statement from .text_perceptron import * in mmdetection/third_party/__init__.py.
Step 3: Compile dependencies of TP model as part of our code is implemented by C++.
>>> cd $OCR_ROOT$/mmdetection/third_party/text_perceptron/
>>> bash tp_setup.shOriginal images can be downloaded from: Total-Text , SCUT-CTW1500.
The formated training datalist and test datalist can be found in demo/text_perceptron_det/datalist/
Modified the paths of imgs/ pretrained_model/ work_space in the config files demo/text_perceptron_det/config/tp_r50_3stages_enlarge.py.
Run the following bash command in the command line,
>>> cd $OCR_ROOT$/demo/text_perceptron_det/
>>> bash train.shDirectly modify the paths (test_dataset, image_prefix, etc.) in the testing script, and run testing:
>>> python test.py Some visualization of detection results are shown:
Reported results on various datasets and trained models download:
| Pretrained | Precision | Recall | Hmean | Links | |
|---|---|---|---|---|---|
| Total-Text (Reported) | SynthText | 88.1 | 78.9 | 83.3 | - |
| Total-Text | SynthText | 85.7 | 81.4 | 83.5 | config, pth (Access Code: vxzn) |
| SCUT-CTW1500 (Reported) | SynthText | 88.7 | 78.2 | 83.1 | - |
| SCUT-CTW1500 | SynthText | 86.1 | 80.0 | 82.9 | config, pth (Access Code: vxzn) |
Note: Models are stored in BaiduYunPan.
If you find this repository is helpful to your research, please feel free to cite us:
@inproceedings{qiao2020text,
title={Text Perceptron: Towards End-to-End Arbitrary-Shaped Text Spotting},
author={Qiao, Liang and Tang, Sanli and Cheng, Zhanzhan and Xu, Yunlu and Niu, Yi and Pu, Shiliang and Wu, Fei},
booktitle={Proceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence (AAAI)},
pages={11899-11907},
year={2020}
}This project is released under the Apache 2.0 license
If there is any suggestion and problem, please feel free to contact the author with qiaoliang6@hikvision.com or chengzhanzhan@hikvision.com.