Awesome
DENet: Disentangled Embedding Network for Visible Watermark Removal
This is official implementation for paper DENet: Disentangled Embedding Network for Visible Watermark Removal [AAAI2023 Oral]
<img src='imgs/framework.png'>Dataset preparation
|--data
|--|--LOGO
|--|--10kmid
|--|--10kgray
|--|--10khigh
Pretrained Model
PSNR | SSIM | LPIPS | |
---|---|---|---|
LOGO-L | 44.24 | 0.9954 | 0.54 |
LOGO-H | 40.83 | 0.9919 | 0.89 |
LOGO-Gray | 42.60 | 0.9944 | 0.53 |
Installation
pip install -r requirements.txt
Training
Train on LOGO-H
bash scripts/train_contrast_attention_on_logo_high.sh
Train on LOGO-L
bash scripts/train_contrast_attention_on_logo_mid.sh
Train on LOGO-Gray
bash scripts/train_contrast_attention_on_logo_gray.sh
Testing
Test on LOGO-H
bash scripts/test_LOGO_10khigh.sh
Test on LOGO-L
bash scripts/test_LOGO_10kmid.sh
Test on LOGO-Gray
bash scripts/test_LOGO_10kgray.sh
Acknowledgement
This code is mainly based on the previous work SLBR.