DarkIR: Robust Low-Light Image Restoration

This repository contains the official model for the paper DarkIR: Robust Low-Light Image Restoration, presented at CVPR 2025.

Daniel Feijoo, Juan C. Benito, Alvaro Garcia, Marcos V. Conde (CIDAUT AI and University of Wuerzburg)

TLDR. Photography in low-light conditions often results in noisy and blurry images. While previous methods address these issues separately, DarkIR proposes the first all-in-one approach for low-light restoration, including illumination, noise, and blur enhancement with a single model.

ABSTRACT

Photography during night or in dark conditions typically suffers from noise, low light and blurring issues due to the dim environment and the common use of long exposure. Although Deblurring and Low-light Image Enhancement (LLIE) are related under these conditions, most approaches in image restoration solve these tasks separately. In this paper, we present an efficient and robust neural network for multi-task low-light image restoration. Instead of following the current tendency of Transformer-based models, we propose new attention mechanisms to enhance the receptive field of efficient CNNs. Our method reduces the computational costs in terms of parameters and MAC operations compared to previous methods. Our model, DarkIR, achieves new state-of-the-art results on the popular LOLBlur, LOLv2 and Real-LOLBlur datasets, being able to generalize on real-world night and dark images.

DarkIR achieves new state-of-the-art results on popular datasets like LOLBlur, LOLv2, and Real-LOLBlur, demonstrating strong generalization capabilities on real-world night and dark images.

Low-light w/ blur	RetinexFormer	DarkIR (ours)
Low-light w/o blur	LEDNet	DarkIR (ours)

 

Network Architecture

Dependencies and Installation

• Python == 3.10.12
• PyTorch == 2.5.1
• CUDA == 12.4
• Other required packages in requirements.txt

# git clone this repository
git clone https://github.com/Fundacion-Cidaut/DarkIR.git
cd DarkIR

# create python environment
python3 -m venv venv_DarkIR
source venv_DarkIR/bin/activate

# install python dependencies
pip install -r requirements.txt

Datasets

The datasets used for training and/or evaluation are:

Dataset	Sets of images	Source
LOL-Blur	10200 training pairs / 1800 test pairs	LEDNet
LOLv2-real	689 training pairs / 100 test pairs	Google Drive
LOLv2-synth	900 training pairs / 100 test pairs	Google Drive
LOL	485 training pairs / 15 test pairs	Official Site
Real-LOLBlur	1354 unpaired images	LEDNet
LSRW-Nikon	3150 training pairs / 20 test pairs	R2RNet
LSRW-Huawei	2450 training pairs / 30 test pairs	R2RNet

You can download each specific dataset and put it on the /data/datasets folder for testing.

Results

We present results in different datasets for DarkIR of different sizes. While DarkIR-m has channel depth of 32, 3.31 M parameters and 7.25 GMACs, DarkIR-l has channel depth 64, 12.96 M parameters and 27.19 GMACs.

Dataset	Model	PSNR	SSIM	LPIPS
LOL-Blur	DarkIR-m	27.00	0.883	0.162
	DarkIR-l	27.30	0.898	0.137
LOLv2-real	DarkIR-m	23.87	0.880	0.186
LOLv2-synth	DarkIR-m	25.54	0.934	0.058
LSRW-Both	DarkIR-m	18.93	0.583	0.412

We present perceptual metrics for Real-LOLBlur dataset:

Model	MUSIQ	NRQM	NIQE
DarkIR-m	48.36	4.983	4.998
DarkIR-l	48.79	4.917	5.051

Evaluation

To check our results you could run the evaluation of DarkIR in each of the datasets:

• Download the weights of the model from OneDrive and put them in /models.
• run python testing.py -p ./options/test/<config.yml>. Default is LOLBlur.

You may also check the qualitative results in Real-LOLBlur and LLIE unpaired by running python testing_unpaired.py -p ./options/test/<config.yml>. Default is RealBlur.

Inference

You can restore a whole set of images in a folder by running:

python inference.py -i <folder_path>

Restored images will be saved in ./images/results.

To inference a video you can run

python inference_video.py -i /path/to/video.mp4

which will be saved in ./videos/results.

Gallery

LOLv2-real

Low-light	SNR-Net	RetinexFormer	DarkIR (ours)	Ground Truth

 

LOLv2-synth

Low-light	SNR-Net	RetinexFormer	DarkIR (ours)	Ground Truth

 

Real-LOLBlur-Night

Citation and acknowledgement

This work has been accepted for publication and presentation at The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2025.

@InProceedings{Feijoo_2025_CVPR,
    author    = {Feijoo, Daniel and Benito, Juan C. and Garcia, Alvaro and Conde, Marcos V.},
    title     = {DarkIR: Robust Low-Light Image Restoration},
    booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {10879-10889}
}

Contact

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