FIRST
Learning-basedEnd-to-end optimizing through differentiable ISP.
Dual-domainRaw denoiser -> noise variations | sRGB denoiser -> ISP variations.
Denoising Paradigm.Trained with synthetic noise, generalize to real-captured images.

DualDn: Dual-domain Denoising via Differentiable ISP

Ruikang Li1,2, Yujin Wang1,†, Shiqi Chen3, Fan Zhang1, Jinwei Gu2, Tianfan Xue2,
1Shanghai AI Laboratory, 2The Chinese University of Hong Kong, 3Zhejiang University
†Indicates Corresponding Author
Arxiv ECCV 2024 Supplementary Code
New BenchmarkWe'll release a new benchmark for dual-denoising.

Trained with Synthetic Noise.
ONE Model. NO Cherry-Picking. NO Fine-Tuning.

SOTA Denoising Model with Exceptional Generalization Ability

Test on UNSEEN smartphone cameras: | | | | | | | |

Duck

Noisy Input

Mask Denoising

Restormer

Original Smartphone ISP

DualDn (ours)

Pressing on keyboard to cycle through different results.

* DualDn NOT trained with any images from these cameras or ISPs.



SOTA Denoising Model with Superior Detail Preservation

Test on PUBLIC DND benchmark: | | | | | | |

0001-18

Noisy Input

Mask Denoising

CycleISP

Restormer

DualDn (ours)

Pressing on keyboard to cycle through different results.

* DND benchmark DOES NOT provide ground truth or training images.



Overview

Due to unpredictable noise characteristics, ALL modern smartphone cameras still struggle with in-the-wild noise.

Previous methods Our DualDn




Poster

BibTeX

@inproceedings{li2024dualdn,
      title={Dualdn: Dual-domain denoising via differentiable isp}, 
      author={Li, Ruikang and Wang, Yujin and Chen, Shiqi and Zhang, Fan and Gu, Jinwei and Xue, Tianfan},
      booktitle={European Conference on Computer Vision},
      pages={160--177},
      year={2025},
      organization={Springer}
}