MIRROR: Multi-scale iterative refinement for robust chinese text recognition

Environment

This work was conducted with PyTorch 1.13.1, CUDA 11.7, python 3.7.

pip install -r requirements.txt

Download

• The Chinese Text Recognition Benchmark datasets are available at the following link: https://github.com/FudanVI/benchmarking-chinese-text-recognition.
• The Traditional Chinese Text Recognition dataset is available at the following link: https://github.com/GitYCC/traditional-chinese-text-recogn-dataset.
• The Chinese Scene Text Recognition Technology Innovation Competition dataset is available at the following link: https://aistudio.baidu.com/datasetdetail/10140.

Training

• Execute the following command to train the MIRROR model.

CUDA_VISIBLE_DEVICES=0 python train.py -c=config/mirror.yml [--model_name=mirror] [--enc_version=base] [--iters=2] [--num_sa_layers=1] [--num_mg_layers=1] [--max_len=32]

Testing

• Execute the following command to test the MIRROR model.

CUDA_VISIBLE_DEVICES=0 python test.py -c=config/mirror.yml --model_name=mirror [--enc_version=base] [--iters=2] [--num_sa_layers=1] [--num_mg_layers=1]