SAIGFormer: A Spatially-Adaptive Illumination-Guided Network for Low-Light Image Enhancement

Framework

Abstract

Recent Transformer-based low-light enhancement methods have made promising progress in recovering global illumination. However, they still struggle with non-uniform lighting scenarios, such as backlit and shadow, appearing as over-exposure or inadequate brightness restoration. To address this challenge, we present a Spatially-Adaptive Illumination-Guided Transformer (SAIGFormer) framework that enables accurate illumination restoration. Specifically, we propose a dynamic integral image representation to model the spatially-varying illumination, and further construct a novel Spatially-Adaptive Integral Illumination Estimator (SAI²E). Moreover, we introduce an Illumination-Guided Multi-head Self-Attention (IG-MSA) mechanism, which leverages the illumination to calibrate the lightness-relevant features toward visual-pleased illumination enhancement. Extensive experiments on five standard low-light datasets and a cross-domain benchmark (LOL-Blur) demonstrate that our SAIGFormer significantly outperforms state-of-the-art methods in both quantitative and qualitative metrics. In particular, our method achieves superior performance in non-uniform illumination enhancement while exhibiting strong generalization capabilities across multiple datasets.

Results and Pretrained Weights

The visualization results of the dataset mentioned in the paper can be downloaded from this link.

The pretrained weights of the relevant dataset can be downloaded from this link.

The extraction code for the above link is: 0909.

Environment Setup

conda create --name saigformer python=3.10
conda activate saigformer
pip install torch==2.6.0 torchvision==0.21.0 --index-url https://download.pytorch.org/whl/cu118
pip install matplotlib scikit-learn scikit-image opencv-python natsort h5py tqdm tensorboard 
pip install torchmetrics thop lmdb numpy pyyaml requests scipy yapf typing triton lpips einops
python setup.py develop --no_cuda_ext

Testing

# LOL_v1
python Enhancement/test_from_dataset.py --opt Options/SAIGFormer_lolv1.yml --weights pretrained_weights/lolv1_pretrained_weight.pth --dataset LOL_v1

# LOL_v2_real
python3 Enhancement/test_from_dataset.py --opt Options/SAIGFormer_lolv2_real.yml --weights pretrained_weights/lolv2_real_pretrained_weight.pth --dataset LOL_v2_real

# LOL_v2_syn
python3 Enhancement/test_from_dataset.py --opt Options/SAIGFormer_lolv2_syn.yml --weights pretrained_weights/lolv2_syn_pretrained_weight.pth --dataset LOL_v2_syn

# SID
python3 Enhancement/test_from_dataset.py --opt Options/SAIGFormer_SID.yml --weights pretrained_weights/sid_pretrained_weight.pth --dataset SID

# SMID
python3 Enhancement/test_from_dataset.py --opt Options/SAIGFormer_SMID.yml --weights pretrained_weights/smid_pretrained_weight.pth --dataset SMID

# LOL_Blur
python3 Enhancement/test_from_dataset.py --opt Options/SAIGFormer_lol_blur.yml --weights pretrained_weights/lol_blur_pretrained_weight.pth --dataset LOL_Blur

Acknowledgments

Thanks to the Retinexformer repository.