metadata
license: mit
tags:
- low-light
- low-light-image-enhancement
- image-enhancement
- image-restoration
- computer-vision
- transformer
- transformers
- vision-transformer
- vision-transformers
- image-segmentation
- illumination
- LoRA
- Mixture of Experts
model-index:
- name: ISALux
results:
- task:
type: low-light-image-enhancement
dataset:
name: LOL-v1
type: LOL-v1
metrics:
- type: PSNR
value: 27.63
name: PSNR
- type: SSIM
value: 0.881
name: SSIM
- task:
type: low-light-image-enhancement
dataset:
name: LOL-v2-Real
type: LOL-v2-Real
metrics:
- type: PSNR
value: 29.76
name: PSNR
- type: SSIM
value: 0.908
name: SSIM
- task:
type: low-light-image-enhancement
dataset:
name: LOL-v2-Synthetic
type: LOL-v2-Synthetic
metrics:
- type: PSNR
value: 30.78
name: PSNR
- type: SSIM
value: 0.956
name: SSIM
- task:
type: low-light-image-enhancement
dataset:
name: SDSD-indoor
type: SDSD-indoor
metrics:
- type: PSNR
value: 30.67
name: PSNR
- type: SSIM
value: 0.909
name: SSIM
- task:
type: low-light-image-enhancement
dataset:
name: SDSD-outdoor
type: SDSD-outdoor
metrics:
- type: PSNR
value: 31.58
name: PSNR
- type: SSIM
value: 0.895
name: SSIM
- task:
type: low-light-image-enhancement
dataset:
name: LOL Blur
type: LOL-Blur
metrics:
- type: PSNR
value: 28.01
name: PSNR
- type: SSIM
value: 0.903
name: SSIM
- task:
type: low-light-image-enhancement
dataset:
name: MEF
type: MEF
metrics:
- type: NIQE
value: 3.58
name: NIQE
- task:
type: low-light-image-enhancement
dataset:
name: LIME
type: LIME
metrics:
- type: NIQE
value: 3.91
name: NIQE
- task:
type: low-light-image-enhancement
dataset:
name: DICM
type: DICM
metrics:
- type: NIQE
value: 3.21
name: NIQE
- task:
type: low-light-image-enhancement
dataset:
name: NPE
type: NPE
metrics:
- type: NIQE
value: 3.4
name: NIQE
pipeline_tag: image-to-image
🌌 ISALux: Illumination & Semantics Aware Transformer with Mixture of Experts
👩💻 Authors:
Raul Balmez, Alexandru Brateanu, Ciprian Orhei, Codruta Ancuti, Cosmin Ancuti
📄 
🔎 Abstract
We introduce ISALux, a novel transformer-based approach for Low-Light Image Enhancement (LLIE) that integrates both illumination and semantic priors.
✨ Key contributions:
- HISA-MSA: A new attention block fusing illumination + semantic segmentation.
- Mixture of Experts (MoE): Improves contextual learning with conditional activation.
- LoRA-enhanced self-attention: Tackles overfitting across diverse light conditions.
Extensive experiments on multiple benchmarks demonstrate state-of-the-art performance.
Ablation studies highlight the role of each proposed component.
🆕 Updates
- 29.07.2025 🎉 Our paper ISALux is live on arXiv!
Dive in to explore methods, results, and ablations. 🚀
📚 Citation
@misc{balmez2025isaluxilluminationsegmentationaware,
title={ISALux: Illumination and Segmentation Aware Transformer Employing Mixture of Experts for Low Light Image Enhancement},
author={Raul Balmez and Alexandru Brateanu and Ciprian Orhei and Codruta Ancuti and Cosmin Ancuti},
year={2025},
eprint={2508.17885},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2508.17885},
}