Project code for ACM MM2020 paper: "TextRay: Contour-based Geometric Modeling for Arbitrary-shaped Scene Text Detection"
- PyTorch
- MMDetection
pip install -r requirements.txt # install dependencies
git submodule update --init # clone mmdetection
cd mmdetection
python setup.py develop # compile mmdetection
cd ../curve/ops/ && sh ./compile.sh # compile custom operations
1, To train a model:
cd experiments/CTW_cheby
./train.sh 0,1,2,3 # training on 4 gpus
Models will be saved in ./work_dirs.
2, To test a model:
cd experiments/CTW_cheby
./test.sh 0,1,2,3 # test on 4 gpus
A .pkl file containing detection results is generated.
Evaluation:
Replace the .pkl filename in curve/notebooks/ctw_eval.ipynb.
The training order is Prertrain_CTW-->CTW_cheby and Pretrain_Total-->Total_cheby.
Models are available in the following links: Google Drive: https://drive.google.com/drive/folders/1nhZuF_S6yvl57RZuC2IgACBqK26sZK5c?usp=sharing Baidu Disk: https://pan.baidu.com/s/1J7fRtKrFdUFvq3eu1P9RVQ (password: cf6b)