SAFL: A Self-Attention Scene Text Recognizer with Focal Loss

This repository implements the SAFL in pytorch.

Installation

conda env create -f environment.yml
conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=10.0 -c pytorch

bash scripts/stn_att_rec.sh

You can test with .lmdb files by

bash scripts/main_test_all.sh

Or test with single image by

bash scripts/main_test_image.sh

We give an example to construct your own datasets. Details please refer to tools/create_svtp_lmdb.py.

If you find this project helpful for your research, please cite the following papers: