MADF-inpainting/Pytorch-MADF

Pytorch-MADF

An Implement of MADF (Image Inpainting by Cascaded Refinement with Mask Awareness). This is a temporally anonymous repo for reproducing results of our submission.

Requirements

• Pytorch 1.0.0
• python 3.*
• numpy 1.13.1

Usages

downlaod repo

• download this repo by the following instruction:

  $ git clone https://github.com/MADF-inpainting/Pytorch-MADF.git
  $ cd Pytorch-MADF-master
  

download datasets

Mask datasets

• Mask datasets can be download here. We train all the datasets with the same mask datasets. Notice that mask images must be 0-255 images.

Places2

• You can download this dataset here. We use the High-resolution version in this repo.

Celeba

• You can download this dataset here. We crop and align the images before training and testing.

Paris street view

• You can download this dataset here.

training

  $ python train.py --train_root train_root --mask_root mask_root --test_root test_root --use_incremental_supervision  

testing

  $ python test.py --list_file test_list --snapshot model_path

Pretrained models

Places2|Celeba|PSV

• All the released models are trained with images of resolution 512 x 512 and the training mask dataset.

Experimental Results

GT	maked input	inpainting result