Face Hallucination via Split-Attention in Split-Attention Network (Accepted by ACM MM 2021)

This repository provides the official PyTorch implementation of the following paper:

Paper link: SISN-MM'21

Requirement

• Python 3.7
• PyTorch >= 1.4.0 (1.5.0 is ok)
• numpy
• skimage
• imageio
• matplotlib
• tqdm

Dataset

Please download FFHQ dataset from here and CelebA dataset from here. After download all datasets, the folder dataset should be like this (take FFHQ as an example):

    dataset    
    └── FFHQ
        ├── 1024X1024
            ├── HR
            └── LR
                ├── X2
                ├── X4
                └── X8
        └── 256X256
            ├── HR
            └── LR
                ├── X2
                ├── X4
                └── X8

Training Model

First, you need to set the necessary parameters in the option.py such as scale, dataset_root, train_val_range, etc. Training the model on the X4 scale as below:

python train.py --model SISN --scale 4

By default, the trained model will be saved in ./pt directory.

Testing model

python test.py --model SISN --scale 4 --pretrain <path_of_pretrained_model> --dataset_root <path_of_input_image> --save_root <path_of_result>

Citation

If you find the code helpful in your resarch or work, please cite the following paper.

@inproceedings{lu2021face,
  title={Face Hallucination via Split-Attention in Split-Attention Network},
  author={Lu, Tao and Wang, Yuanzhi and Zhang, Yanduo and Wang, Yu and Wei, Liu and Wang, Zhongyuan and Jiang, Junjun},
  booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
  pages={5501--5509},
  year={2021}
}