This code is the official implementation of ICCV 2019 paper "Mop Moire Patterns Using Mopnet".

1. Linux
2. python2 or 3
3. NVIDIA GPU + CUDA CuDNN (CUDA 8.0)

1. Install PyTorch from https://pytorch.org
2. Install Torch vision from https://github.com/pytorch/vision
3. Install python package: numpy, scipy, PIL, math, skimage, visdom

1. VGG16 https://drive.google.com/open?id=1wNHZOyTr3veCHU-JaQwmSV7JbKWIMbAT
2. classifier https://drive.google.com/drive/folders/1MkSVkzwWeKmaIRIzq9-J4ZINEFq4TRFA
3. caorse pre-trained edge predictor https://drive.google.com/drive/folders/1MkSVkzwWeKmaIRIzq9-J4ZINEFq4TRFA
4. totally pre-trained mopnet https://drive.google.com/drive/folders/1MkSVkzwWeKmaIRIzq9-J4ZINEFq4TRFA

Download totally pre-trained mopnet and classifier. Put color_epoch_95.pth and geo_epoch_95.pth into folder classifier. Put netEdge_epoch_150.pth and netG_epoch_150.pth into folder mopnet. Download testset from https://drive.google.com/open?id=1a-4iwy3ujCfC8llBaimjXnVfOM9oGKAV

Change the dataroot in run_test.sh.

Create folders: mkdir results mkdir results/d mkdir results/o mkdir results/g

Execute bash run_test.sh Then you will get moire free images. For fair comparison, we compute the PNSR and SSIM in Matlab which is the same as TIP18. (Moire Photo Restoration Using Multiresolution Convolutional Neural Networks) So you can run matlab test_with_matlabcode.m to get quantitative results.

Download caorse pre-trained edge predicotr and put it into folder edge. Download VGG and put it into folder models. Download the dataset from https://drive.google.com/open?id=1a-4iwy3ujCfC8llBaimjXnVfOM9oGKAV The whole benchmark training set please contact the author of TIP18. Change the dataroot and valDataroot in run_train.sh. Open the visualization: python -m visdom.server -port 8098

Execute bash run_train.sh

@inproceedings{he2019mop,
  title={Mop Moire Patterns Using MopNet},
  author={He, Bin and Wang, Ce and Shi, Boxin and Duan, Ling-Yu},
  booktitle=ICCV,
  pages={2424--2432},
  year={2019}
}

If you have any question, feel free to concat me with [email protected].