WatermarkRemoval-WDNet-WACV2021

Related tags

Deep Learning WDNet
Overview

WatermarkRemoval-WDNet-WACV2021

Thank you for your attention.

Citation

Please cite the related works in your publications if it helps your research:

@InProceedings{Liu_2021_WACV,
author = {Yang Liu and
          Zhen Zhu and
          Xiang Bai},
title = {WDNet: Watermark-Decomposition Network for Visible Watermark Removal},
booktitle = { 2021 {IEEE/CVF} Winter Conference on Applications of Computer Vision (WACV)},
publisher = {{IEEE}},
page = {3685-3693},
year = {2021}
}

Dataset CLWD

CLWD

Pretraied Model

Thanks for the help of @ChaiHuanhuan, who trained the WDNet and provided a pretrained WDNet model. This model is trained for 50 epoches.

You might also like...
Comments
  • Pretrained Model

    Pretrained Model

    Congratulations for the excellent work!

    Because I don't have computational capacity, could you provide the pre-trained model in order to do some tests?

    opened by joaocps 6
  • About PSNR Evaluation

    About PSNR Evaluation

    Hi, thanks for sharing your source code and CLWD dataset. I train your model from scratch on CLWD dataset. When I test the well-trained model, I find something wrong with your evaluation code. The default return format of imread in opencv is unsigned int8, so that it will lead to numeric overflow problem when you compute the psnr directly, in which it involves the subtraction operation. I think it should be converted to int32 or float32 before computing the psnr score.

    opened by jimleungjing 0
  • vgg l1 loss

    vgg l1 loss

    In vgg.py, as the forward, will return 4 outputs. But in WDNet.py line252-line254, only use top3 outputs to compute the l1 loss

    And in paper, To calculate L1 perceptual loss, we use the outputs of the relu2 2 layer of a pre-trained VGG-16 network [25] to represent the learned features of Y^o and Y and then compute their L1 difference. But in this project, compute the MSELoss

    opened by luoling1993 0
Owner
LUYI
LUYI
GitHub