Home

Awesome

Nested Deformable Multi-head Attention for Facial Image Inpainting [WACV-23]

PyTorch implementation of Nested Deformable Multi-head Attention for Facial Image Inpainting.

<p align="center"><img src="architecture.png" alt="SPL" width="60%"></p>

Prerequisites

Usage

Keep your dataset by placing images like:

dataset
├── celeb
│   ├── input
│   │   ├── 1.png 
│   │   ├── 2.png 
│   │   └── ...
│   ├── mask
│   │   ├── 1.png 
│   │   ├── 2.png 
│   │   └── ...    
│   └── target
│       ├── 1.png
│       └── ...

Checkpoints:

Download the checkpoint: Checkpoints

The checkpoints are provided for:
    CelebA-HQ dataset
    Keep the checkpoints in "./checkpoints/"

To test the network:

python test.py
    

The results will be stored in:

./results

Citation

If our method is useful for your research, please consider citing:

@inproceedings{phutke2023nested,
  title={Nested Deformable Multi-head Attention for Facial Image Inpainting},
  author={Phutke, Shruti S and Murala, Subrahmanyam},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},  pages={6078--6087},
  year={2023}
}