Blind Image Inpainting via Omni-dimensional Gated Attention and Wavelet Queries [CVPRW-23]

PyTorch implementation of Blind Image Inpainting via Omni-dimensional Gated Attention and Wavelet Queries.

Prerequisites

• Python 3.6+
• PyTorch>1.0
• cv2, numpy, PIL, torchvision

Usage

Keep your dataset by placing images like:

datasets
├── celeb
│   ├── input
│   │   ├── 1.png 
│   │   ├── 2.png 
│   │   └── ...
│   └── target
│       ├── 1.png
│       └── ...
├── ffhq
│   ├── input
│   │   ├── 1.png 
│   │   ├── 2.png 
│   │   └── ...
│   └── target
│       ├── 1.png 
│       └── ...
├── paris
│   ├── input
│   │   ├── 1.png 
│   │   ├── 2.png 
│   │   └── ...
│   └── target
│       ├── 1.png 
│       └── ...
└── places2
    ├── input
    │   ├── 1.png 
    │   ├── 2.png 
    │   └── ...
    └── target
        ├── 1.png 
        └── ...

Dataset Generation

The code for generating the training and testing images is provided in "dataset_generation.py"

Checkpoints:

The checkpoints are provided for:
1. CelebA-HQ dataset.
2. FFHQ dataset.
3. Paris Street View dataset.
4. Places2 dataset.
4. Keep the checkpoints in "./checkpoints/dataset.pth"

Download the checkpoint:

To test the network:

CelebA_HQ dataset:
    python test.py --data_path_test ./datasets/celeb/ --task_name inpaint --dataset_name celeb --model_file ./checkpoints/celeb.pth

FFHQ dataset:
    python test.py --data_path_test ./datasets/ffhq/ --task_name inpaint --dataset_name ffhq --model_file ./checkpoints/ffhq.pth

Places2 dataset:
    python test.py --data_path_test ./datasets/places/ --task_name inpaint --dataset_name places --model_file ./checkpoints/places.pth

Paris Street View dataset:
    python test.py --data_path_test ./datasets/paris/ --task_name inpaint --dataset_name paris --model_file ./checkpoints/paris.pth 

The results will be stored in:

./results/respective_dataset_name/

Citation

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

@inproceedings{phutke2023blind,
  title={Blind Image Inpainting via Omni-Dimensional Gated Attention and Wavelet Queries},
  author={Phutke, Shruti S and Kulkarni, Ashutosh and Vipparthi, Santosh Kumar and Murala, Subrahmanyam},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={1251--1260},
  year={2023}
}