SymmFCNet

Torch implementation for Learning Symmetry Consistent Deep CNNs for Face Completion

(New)Pytorch Version can be found here.

+ Please note that there are some different implementations between torch and pytorch version. 
+ Please refer to pytorch version (https://github.com/csxmli2016/SymmFCNet_pytorch).

SymmFCNet framework (Torch Version)

Overview of our SymmFCNet. Red, green and blue lines represent the pixel-wise correspondence between the input and the flip image.

Red: missing pixels (input) to non-occluded pixels (flip);
Green: missing pixels (input) to missing pixels (flip);
Blue: remaining pixels (input) to remaining pixels (flip).

Models

Download the pre-trained model with the following url and put it into ./checkpoints/.

Testing

th test.lua

Completion results

Type	Input	SymmFCNet	Ground-truth
Regular Mask
Irregular Mask
Real Image

Requirements and Dependencies

Torch
Cuda-8.0
Stn

Citation

@article{li2018learning,
  title={Learning Symmetry Consistent Deep CNNs for Face Completion},
  author={Li, Xiaoming and Liu, Ming and Zhu, Jieru and Zuo, Wangmeng and Wang, Meng and Hu, Guosheng and Zhang, Lei},
  journal={arXiv preprint arXiv:1812.07741},
  year={2018}
}