CentripetalText

Codebase for NeurIPS2021 "CentripetalText: An Efficient Text Instance Representation for Scene Text Detection" [Paper link]

Recommended environment

Python 3.6+
Pytorch 1.1.0
torchvision 0.3
mmcv 0.2.12
editdistance
Polygon3
pyclipper
opencv-python 3.4.2.17
Cython
numpy
json

Dataset

CentripetalText
└── data
    ├── total_text
    │   ├── Images
    │   │   ├── Train
    │   │   └── Test
    │   └── Groundtruth
    │       └── Polygon
    │           ├── Train
    │           └── Test
    ├── MSRA-TD500
    │   ├── train
    │   └── test
    ├── HUST-TR400
    └── SynthText
        ├── 1
        ├── ...
        ├── 200
        └── gt.mat

Training

CUDA_VISIBLE_DEVICES=0,1,2,3 python train.py ${CONFIG_FILE}

For example:

CUDA_VISIBLE_DEVICES=0,1,2,3 python train.py config/ct/ct_r18_tt.py

Test

Pre-trained checkpoints can be downloaded from baidu cloud (Password: 31mk). When testing, move the file to checkpoints/.

python test.py ${CONFIG_FILE} ${CHECKPOINT_FILE}

For example:

python test.py config/ct/ct_r18_tt.py checkpoints/ct_r18_tt/checkpoint.pth.tar

Evaluation

cd eval/
./eval_tt.sh

TODO

• [x] Release code
• [x] Trained models
• [ ] Recognition codes

Citation

@inproceedings{sheng2021centripetaltext,
    title={CentripetalText: An Efficient Text Instance Representation for Scene Text Detection},
    author={Tao Sheng and Jie Chen and Zhouhui Lian},
    booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
    year={2021}
}