mrfisher97 / dewords

百度网盘AI大赛——图像处理挑战赛：手写文字擦除第5名方案

This repository is the implementation of PaddlePaddle-based handwriting removal model.

File Structure

├── dataloader.py
├── Dataset
├── generate_files.py
├── Log
├── loss.py
├── main.py
├── model.py
├── README.md
├── requirement.txt
├── run.sh
└── utils.py

Environment

The environment can be seen in requirement.txt file.

Generate file list

Running the generate_files.py to generate file list in dataset directory.

Training

Once the data is well prepared, you can begin training:

python train_segformer.py\
    --train 1\
    --arch segformer_b2\
    --dataRoot 'Your Path'\

Generate

If you want to predict the results, run:

python train_segformer.py\
    --loadSize 512\
    --train 0\
    --arch segformer_b2\
    --testDataRoot 'Your Path'\
    --modelLog 'Your Path'