README.md

C3N：Content-Constrained Convolutional Network for Mural Image Completion

Paper

Requirements

numpy==1.14.4

Pillow==5.1.0

six==1.11.0

tensorboardX==1.2

torch==0.4.1

torchvision==0.2.1

tqdm==4.23.4

Preparation works

To generate binary masks, use

python generate_data.py

To generate the image covered by the mask, that is, generate the simulated damaged image, use

python 1test.py

Training and testing

To conduct network model training, use

python train.py

The image data set and mask data set can be simply modified at the beginning of the code as required.

To generate a repair image, use

python 2test.py

Citation

If you find our code or paper useful, please cite the paper:

@article{PengWZ23,
title = {C3N: Content-constrained convolutional network for mural image completion},
author = {Xianlin Peng, Huayu Zhao, Xiaoyu Wang, Yongqin Zhang, Zhan Li, Qunxi Zhang, Jun Wang, Jinye Peng, Haida Liang},
journal = {Neural Computing and Applications},
volume = {35},
article id = {},
pages = {1959-1970},
year={2023}
}