This code repository contains an implementation of EAST Text detector. EAST: An Efficient and Accurate Scene Text Detector (CVPR 2017).
Original images can be downloaded from: ICDAR2015.
The formatted training datalist and test datalist can be found in demo/text_detection/datalist/
Modified the paths ("imgs"/ "pretrained_model"/ "work_space", etc.) in the config files demo/text_detection/east/config/east_r50_rbox.py.
Run the following bash command in the command line,
cd $DAVAR_LAB_OCR_ROOT$/demo/text_detection/east/
bash dist_train.shWe provide the implementation of online validation. If you want to close it to save training time, you may modify the startup script to add
--no-validatecommand.
We provide a demo of forward inference and visualization. You can modify the paths (test_dataset, image_prefix, etc.) in the testing script, and start testing:
python test.py Some visualization of detection results are shown:
The general evaluation tool can be found in davarocr/demo/text_detection/evaluation/.
All of the models are re-implemented and well trained in the based on the opensourced framework mmdetection.
Results on various datasets and trained models download:
| Dataset | Mode | Backbone | Pre-trained | Test Scale | Precision | Recall | Hmean | Links |
|---|---|---|---|---|---|---|---|---|
| ICDAR2015 (Reported) | RBOX | PAVNET2x | ImageNet | (L-1280, S-720) | 83.57 | 73.47 | 78.20 | |
| ICDAR2015 | RBOX | ResNet-50 | ImageNet | (L-1280, S-720) | 80.53 | 79.44 | 79.98 | config, pth (Access Code: 7968) |
| ICDAR2015 | RBOX | ResNet-50 | ImageNet | (L-1600, S-900) | 79.45 | 80.4 | 79.92 | config ,pth (Access Code: 7968) |
| ICDAR2015 (Reported) | QUAD | PAVNET2x | ImageNet | (L-1600, S-900) | 80.18 | 74.19 | 77.07 | |
| ICDAR2015 | QUAD | ResNet-50 | ImageNet | (L-1280, S-720) | 71.43 | 79.20 | 75.11 | config , pth (Access Code: 0D9h) |
| ICDAR2015 | QUAD | ResNet-50 | ImageNet | (L-1600, S-900) | 71.52 | 79.30 | 75.21 | config , pth (Access Code: 0D9h) |
** Note **: Above links are from BaiduYunPan. Models can also be downloaded from Google Drive
@inproceedings{Zhou_2017,
title={EAST: An Efficient and Acccurate Scene Text Detector},
author={Xinyu Zhou and Cong Yao and He Wen and Yuzhi Wang and Shuchang Zhou and Weiran He and Jiajun Liang},
booktitle={2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2017}
}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.