This repository is the official implementation of "PSSTRNET: PROGRESSIVE SEGMENTATION-GUIDED SCENE TEXT REMOVAL NETWORK"

The pretrained models on SCUT-Syn and SCUT-EnsText. [BaiduDisk] [GoogleDrive]

The visualization results on SCUT-Syn and SCUT-EnsText. [BaiduDisk] [GoogleDrive]

The evaluation metrics can refer the EraseNet. EraseNet

Inference

0. Create a virtual env:

conda create -n psstr pyhton=3.10 -y

conda activate psstr

1. Install requirements:

pip install -r requirements.txt

2. Run the command:

python inference.py -i /path/to/your/images

NOTE: you can run with one or several images.