In recent years, the field of image inpainting has developed rapidly, but most deep methods are strongly tied to the resolution of the images on which they were trained. A slight resolution increase leads to artifacts and unsatisfactory filling quality. These methods are therefore unsuitable for interactive image processing. In this article, we propose a method that solves the problem of inpainting arbitrary-size images. We also describe a way to better restore texture fragments in the filled area. Moreover, this approach can work with existing inpainting models, making them resolution independent. We also created a GIMP plugin that implements our technique.
- Python 3.7
- Install requirements with
pip install -r requirements.txt
- Download weights and save it in
weightsfolder. - Put your images as shown in
Test/Inputs - Run:
python test.py
You can find all the images involved in our comparison here
- GIMP 2.10
- Ubuntu 18.04 LTS
- macOS Mojave 10.14.6
- Open GIMP and go to Preferences -> Folders -> Plug-ins, add the folder
gimp-pluginsfrom this repo and close GIMP. - Download weights and save it in
gimp-plugins/Inpainting/weightsfolder. - Open terminal and run:
bash installGimpML.sh - Open GIMP.
You can find example of usage: youtube.
Please note that the mask must be exactly binary. Otherwise the filling result will be terrible.
To do this, remove antialiasing in selection:
@article{Moskalenko_2020,
doi = {10.51130/graphicon-2020-2-4-18},
url = {https://doi.org/10.51130%2Fgraphicon-2020-2-4-18},
year = 2020,
month = {dec},
pages = {short18--1--short18--9},
author = {Andrey Moskalenko and Mikhail Erofeev and Dmitriy Vatolin},
title = {Deep Two-Stage High-Resolution Image Inpainting},
journal = {Proceedings of the 30th International Conference on Computer Graphics and Machine Vision ({GraphiCon} 2020). Part 2}
}
We are largely benefiting from:
[1] https://github.com/hughplay/DFNet
[2] https://github.com/kritiksoman/GIMP-ML/