Restorable Image Operators with Quasi-Invertible Networks

This is the PyTorch implementation of paper: Restorable Image Operators with Quasi-Invertible Networks (AAAI 2022)..

Brief Introduction

We propose a quasi-invertible model that learns common image processing operators in a restorable fashion: the learned image operators can generate visually pleasing results with the original content embedded.

Dependencies

• Python 3
• PyTorch >= 1.0
• NVIDIA GPU + CUDA
• Python packages: pip install numpy opencv-python lmdb pyyaml

Dataset Preparation

We use Adobe5K dataset for training and evaluation.

Get Started

Training and testing codes are in 'codes/'. Refer to the training scripts for the detailed options.

Acknowledgement

The code is based on invertible image rescaling. Thanks the authors for sharing their code.

Contact

If you have any questions, please contact [email protected].

Citation

If you find the code useful please cite:

@inproceedings{ouyang2022restorable,
  title={Restorable Image Operators with Quasi-Invertible Networks},
  author={Ouyang, Hao and Wang, Tengfei and Chen, Qifeng},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={36},
  number={2},
  pages={2008--2016},
  year={2022}
}