PyTorch Partial Convolution Inpainting

This is the Re-implementation of "Image Inpainting for Irregular Holes Using Partial Convolutions".

This is NOT an official implementation by the author.

Result

From top to bottom, The input image, The mask image, The raw output, The outpu with ground truth except for mask area, and the ground truth image.

Setup

Clone this repo

git clone https://github.com/tanimutomo/partialconv.git

Install requirements

The required libraries are written in ./requirements.txt.

Quick Run

Download the pretrained model

You can donwload the pretrained model which is the same as the model output the above images.
Please put the pretrained model file to partialconv/.

Run

• Quick Run

python predict.py

• You can specify an image / a mask / a model paths

python predict.py --img <image_path> --mask <mask_path> --model <model_path>

Train and Test

• Copy the config yaml file.

cd partialconv/
cp default-config.yml config.yml

• Customize config.yml
• Run main.py

Information

Original Paper

• Title
  • Image Inpainting for Irregular Holes Using Partial Convolutions
• Author
  • Guilin Liu / Fitsum A. Reda / Kevin J. Shih / Ting-Chun Wang / Andrew Tao / Bryan Catanzaro
• Affiliation
  • NVIDIA Corporation
• Official Implementation
  • https://github.com/NVIDIA/partialconv
• Other Information
  • http://masc.cs.gmu.edu/wiki/partialconv

Reference Codes

I refered the following repositories for this implementation. This implementation integrate the strong points of these refered codes.

• https://github.com/NVIDIA/partialconv
• https://github.com/MathiasGruber/PConv-Keras
• https://github.com/naoto0804/pytorch-inpainting-with-partial-conv/