Retinex-MEF

Codes for Retinex-MEF: Retinex-based Glare Effects Aware Unsupervised Multi-Exposure Image Fusion (ICCV 2025)

Haowen Bai, Jiangshe Zhang, Zixiang Zhao, Lilun Deng, Yukun Cui, Shuang Xu.

-[Paper]

-[ArXiv]

Abstract

Multi-exposure image fusion (MEF) synthesizes multiple, differently exposed images of the same scene into a single, well-exposed composite. Retinex theory, which separates image illumination from scene reflectance, provides a natural framework to ensure consistent scene representation and effective information fusion across varied exposure levels. However, the conventional pixel-wise multiplication of illumination and reflectance inadequately models the glare effect induced by overexposure. To address this limitation, we introduce an unsupervised and controllable method termed Retinex-MEF. Specifically, our method decomposes multi-exposure images into separate illumination components with a shared reflectance component, and effectively models the glare induced by overexposure. The shared reflectance is learned via a bidirectional loss, which enables our approach to effectively mitigate the glare effect. Furthermore, we introduce a controllable exposure fusion criterion, enabling global exposure adjustments while preserving contrast, thus overcoming the constraints of a fixed exposure level. Extensive experiments on diverse datasets, including underexposure-overexposure fusion, exposure controlled fusion, and homogeneous extreme exposure fusion, demonstrate the effective decomposition and flexible fusion capability of our model

Update

• [2025/9] Release the code.

Citation

@inproceedings{bai2024retinex,
  title={Retinex-MEF: Retinex-based Glare Effects Aware Unsupervised Multi-Exposure Image Fusion},
  author={Bai, Haowen and Zhang, Jiangshe and Zhao, Zixiang and Deng, Lilun and Cui, Yukun and Xu, Shuang},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  year={2025}
}

🌐 Usage

🏊 Training

1. Data Preparation

Modify 'raw_path' in './data/SICE_arrange.py' to the path for the raw SICE training set.

run

python ./data/SICE_arrange.py

to sort the data by exposure from low to high.

run

python ./data/SICE_processing.py

to extract patches from the sorted data. The cropped and processed data will be stored in './data/SICE_training'. Where img_1 is the underexposed image, img_2 is the overexposed image, img_3 is other images from the sequence with different exposure levels.

Here is a sample dataset containing 10 patches. (Note: In this example, img_3 is the same as img_1.)

2. Commence Training

run

python train.py

The models will be saved in the ./exp directory based on the timestamp.

🏄 Testing

1. Pretrained models

Pretrained models are available in ./model/

2. Test cases

The 'test_cases' folder contains four examples that appear in the main paper. Running

python test.py

will fuse these cases, and the fusion results will be saved in the 'test_results' folder.

3. Test customization

Modify the variables in test.py as needed:

'path_model' (the path of the pretrained model),

'path_img1' (the path of the under-exposure images),

'path_img2' (the path of the over-exposure images),

'path_result' (the path to save the fusion result).

Then run

python test.py

and the fusion results will be saved in the 'path_result'.