Image Inpainting with External-internal Learning and Monochromic Bottleneck

This repository is for the CVPR 2021 paper: 'Image Inpainting with External-internal Learning and Monochromic Bottleneck'

paper | project website

Introduction

The proposed method can be applied to improve the color consistency of leaning-based image inpainting results. The progressive internal color propagation shows strong performance even with large mask ratios.

Prerequisites

• Python 3.6
• Pytorch 1.6
• Numpy

Installation

git clone https://github.com/Tengfei-Wang/external-internal-inpainting.git
cd external-internal-inpainting

Quick Start

To try our internal colorization method:

python main.py  --img_path images/input2.png --gray_path images/gray2.png  --mask_path images/mask2.png  --pyramid_height 3

The colorization results are placed in ./results.

Citation

If you find this work useful for your research, please cite:

@InProceedings{wang2021image,
     title={Image Inpainting with External-internal Learning and Monochromic Bottleneck}, 
     author={Tengfei Wang, Hao Ouyang and Qifeng Chen},
     booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
     year = {2021}
}             

Contact

Please send emails to [email protected] if there is any question