This is the project page of 《Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training》, which is accepted by NeurIPS 2021. [paper] [supplementary]
The quantitative results of models finetuned with the image pairs generated by our PNGAN are listed in the following table. Our models achieve state-of-the-art results.
The qualitative comparisons of noise generation between previous methods and our PNGAN are shown in the following figure. Our PNGAN creates more realistic and visually natural noisy images.
We provide the denoised results on DND benchmark, see folder 'DND_results'. Our method ranks the 4th place on the public leaderboard.
@inproceedings{cai2021learning,
title={Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training},
author={Cai, Yuanhao and Hu, Xiaowan and Wang, Haoqian and Zhang, Yulun and Pfister, Hanspeter and Wei, Donglai},
booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
year={2021}}