CVPR Findings2026
RodNet: Visual Pathway-Inspired Adaptive Sparse Network for Efficient Low-Light Image Enhancement
Boheng Liu, Ziyu Li, Zhong Zhang, Mengrui Xu, Chenghua Duan, Dehao Liu, Qing Li, Xia Wu
Abstract
Low-light image enhancement (LLIE) requires real-time performance on resource-constrained edge devices, yet existing methods, despite achieving compact model sizes and superior enhancement quality, still suffer from high computational complexity that limits practical deployment in time-critical applications. Inspired by the asymmetric organization of retinal rod-cone photoreceptors and cortical sparse coding principles, we propose RodNet for efficient LLIE. RodNet introduces three key innovations: (1) a Rod-Cone Dual Pathway with asymmetric channel allocation that processes intensity with fewer channels than chromatic information, inspired by biological photoreceptor organization; (2) an Interleaved Multi-scale Cross Attention (IMCA) module that unifies bidirectional feature exchange in a single forward pass with contrast-conditioned modulation; and (3) a Contrast-Aware Dynamic Channel Pruning (CDCP) mechanism that adaptively modulates channel activation based on local Weber contrast, emulating cortical sparse coding. Extensive experiments demonstrate that RodNet achieves state-of-the-art performance across multiple benchmark datasets, while reducing computational cost by 48% with only 1.71M parameters and 4.21 GFLOPs, enabling practical real-time deployment on mobile and edge computing platforms.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Learning a Simple Low-Light Image Enhancer from Paired Low-Light Instances cites SNR-Aware Low-light Image Enhancement
- Light the Night: A Multi-Condition Diffusion Framework for Unpaired Low-Light Enhancement in Autonomous Driving cites SNR-Aware Low-light Image Enhancement
- DarkIR: Robust Low-Light Image Restoration cites SNR-Aware Low-light Image Enhancement
- DarkIR: Robust Low-Light Image Restoration cites Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement
- Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement cites SNR-Aware Low-light Image Enhancement
Builds on6
- Retinexformer: One-stage Retinex-based Transformer for Low-light Image EnhancementYuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang et al.ICCV 2023 · 615 citations
- SNR-Aware Low-light Image EnhancementXiaogang Xu, Ruixing Wang, Chi-Wing Fu, Jiaya JiaCVPR 2022 · 552 citations
- Learning a Simple Low-Light Image Enhancer from Paired Low-Light InstancesZhenqi Fu, Yan Yang, Xiaotong Tu, Yue Huang et al.CVPR 2023
- Light the Night: A Multi-Condition Diffusion Framework for Unpaired Low-Light Enhancement in Autonomous DrivingJinlong Li, Baolu Li, Zhengzhong Tu, Xinyu Liu et al.CVPR 2024
- DarkIR: Robust Low-Light Image RestorationDaniel Feijoo, Juan C. Benito, Álvaro García, Marcos V. CondeCVPR 2025
Related papers
- MobileIE: An Extremely Lightweight and Effective ConvNet for Real-Time Image Enhancement on Mobile DevicesHailong Yan, Ao Li, Xiangtao Zhang, Zhe Liu et al.ICCV 2025 · 12 citations
- Multinex: Lightweight Low-light Image Enhancement via Multi-prior RetinexAlexandru Brateanu, Tingting Mu, Codruta O. Ancuti, Cosmin AncutiCVPR 2026 · 10 citations
- Spike-RetinexFormer: Rethinking Low-light Image Enhancement with Spiking Neural NetworksHongzhi Wang, Xiubo Liang, Jinxing Han, Weidong GengNeurIPS 2025
- Retinex-Inspired Unrolling With Cooperative Prior Architecture Search for Low-Light Image EnhancementRisheng Liu, Long Ma, Jiaao Zhang, Xin Fan et al.CVPR 2021
- Integrating Semantic Segmentation and Retinex Model for Low-Light Image EnhancementMinhao Fan, Wenjing Wang, Wenhan Yang, Jiaying LiuACM MM 2020 · 135 citations
