Official PyTorch implementation of CAM

This official pytorch implementation for "Class-Aware Mask-guided feature refinement for scene text recognition".

Data Preparation

Data	path
Synthetic Text Data	SynthText, Synth90k (Baiduyun with passwd: wi05)
Scene Text Recognition Benchmarks	IIIT5k, SVT, IC13, IC15, SVTP, CUTE, COCOText, CTW, Total-Text, HOST, WOST,WordArt

Checkpoint

You can download different version of our model in the link below.

Base

Nano

Tiny

Setup

conda create -n cam python=3.8 -y
conda activate cam
pip install -r requirements.txt

Run

Train
```
bash script/train.sh
```
Evaluation
- modify the script/eval_tiny.sh ,change the ''path_to_pth'' to your own path, which is similar to eval_nano.sh and eval_base.sh
```
bash script/eval_tiny.sh
```

Result

Citation

@article{yang2024class,
  title={Class-Aware Mask-guided feature refinement for scene text recognition},
  author={Yang, Mingkun and Yang, Biao and Liao, Minghui and Zhu, Yingying and Bai, Xiang},
  journal={Pattern Recognition},
  volume={149},
  pages={110244},
  year={2024},
  publisher={Elsevier}
}

This repository is built upon ConvNeXt-V2, thanks very much!