This is the official implementation of the paper "Rethinking Reconstruction and Denoising in the Dark:New Perspective, General Architecture and Beyond".

Comparison among recent state-of-the-art methods and our method.

Recently, enhancing image quality in the original RAW domain has garnered significant attention, with denoising and reconstruction emerging as fundamental tasks. Although some works attempt to couple these tasks, they primarily focus on multi-stage learning while neglecting task associativity within a broader parameter space, leading to suboptimal performance. This work introduces a novel approach by rethinking denoising and reconstruction from a “backbone-head” perspective, leveraging the stronger shared parameter space offered by the backbone, compared to the encoder used in existing works. We derive task specific heads with fewer parameters to mitigate learning pressure. By incorporating chromaticity-aware attention into the backbone and introducing an adaptive denoising prior during training, we enable simultaneous reconstruction and denoising. Additionally, we design a dual-head interaction module to capture the latent correspondence between the two tasks, significantly enhancing multi-task accuracy. Extensive experiments validate the superiority of the proposed method.

1.Clone Repo

git clone https://github.com/csmty/CANS.git

2.Create Conda Environment and install Dependencies

conda create -n CANS python=3.10
conda activate CANS
pip install -r requirements.txt -f https://download.pytorch.org/whl/cu121/torch_stable.html

1.You can change the input dataset path and the output path by modifying the paths in the configs/CANS_Plus/Sony.yaml file.

2.You can perform training and testing through the following commands: sh train.sh, sh test.sh.

If you use our code and dataset for research, please cite our paper:

@inproceedings{ma2025rethinking,
  title={Rethinking Reconstruction and Denoising in the Dark: New Perspective, General Architecture and Beyond},
  author={Ma, Tengyu and Ma, Long and Li, Ziye and Wang, Yuetong and Liu, Jinyuan and Xu, Chengpei and Liu, Risheng},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={2323--2332},
  year={2025}
}

If you have any questions or suggestions about this repo, please feel free to contact me (tyma0913@gmail.com).

This repository borrows heavily from DNF. Thanks for their outstanding contributions.