MambaLIE: Scene Light Intensity-Boosted Low-Light Image Enhancement with State Space Model

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision, Emerging Technologies & Innovation · Depth: Expert, extended

Summary

MambaLIE is a novel low-light image enhancement (LIE) method based on a State Space Model (SSM) that addresses the limitations of Convolutional Neural Networks (CNNs) and Transformers. It introduces a scene light intensity prior to improve the structural distribution of illumination and proposes a Locally Enhanced State Space Model (LESSM) for efficient light enhancement. LESSM combines an SSM branch for modeling long-range dependencies with linear time complexity and a local enhanced branch for local feature representations. MambaLIE outperforms existing CNN-based and Transformer-based LIE methods on four widely used synthetic benchmarks (LOLv1, LOLv2-real, LOLv2-syn, MIT-Adobe FiveK) and five publicly available real-world benchmarks (LIME, MEF, NPE, DICM, VV) in terms of accuracy, speed, and model size. The model is trained on 128x128 image patches with a batch size of 12, using a 4-level encoder-decoder architecture with {1,2,4,8} GLEMs, and optimized with Adam for 1000-4000 epochs on an NVIDIA RTX 4090 GPU.

Key takeaway

For Machine Learning Engineers developing low-light image processing solutions for consumer electronics, MambaLIE offers a superior balance of performance and efficiency. You should consider integrating its State Space Model-based architecture to achieve high-quality enhancement with linear computational complexity, making it suitable for deployment on resource-constrained devices like mobile phones and digital cameras. This approach can significantly improve downstream vision tasks such as object detection.

Key insights

MambaLIE uses a State Space Model and scene light intensity prior for efficient, high-quality low-light image enhancement.

Principles

• Combine global SSM with local enhancement for optimal feature learning.
• Scene light intensity prior improves structural illumination distribution.
• Linear time complexity SSMs are efficient for long-range dependencies.

Method

MambaLIE computes a scene light intensity prior via a mean filter, gates it with the low-light input, then processes it through a U-shaped encoder-decoder framework using Locally Enhanced State Space Models (LESSM), channel attention, and a Dual Gated Feed-Forward Network.

In practice

• Deploy MambaLIE on resource-constrained consumer devices.
• Use scene light intensity prior to guide image enhancement.
• Integrate bidirectional SSMs for robust context modeling.

Topics

• Low-Light Image Enhancement
• State Space Models
• Mamba Architecture
• Computer Vision
• Image Restoration
• Consumer Electronics

Code references

• ghfkahfk/MambaLIEcode

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Spatial-Aware Reduction Framework: Towards Efficient and Faithful Visual State Space Models — Token reduction in visual State Space Models requires spatial awareness to prevent performance collapse.
• Multinex: An ultra lightweight AI model advancing low light image enhancement — Multinex uses classical color vision and Retinex principles for efficient, compact low-light image enhancement.
• ConvVitMamba: Efficient Multiscale Convolution, Transformer, and Mamba-Based Sequence modelling for Hyperspectral Image Classification — ConvVitMamba efficiently classifies hyperspectral images by combining multiscale CNNs, Vision Transformers, and Mamba-inspired sequence mixing.
• MamBOA: State-Space Architecture for Video Recognition — MamBOA recasts selective state-space recurrence as a native motion synthesizer for robust video action recognition.
• DM3D: Dynamic Mamba via Offset-Guided Feature Resampling for Point Cloud Understanding — Dynamic, learnable serialization via deformable Mamba enhances point cloud understanding by adapting to geometric structures.
• Reload-Mamba: Hierarchical Anti-Dilution State-Space Modeling for Multi-Class Semantic Segmentation — Reload-Mamba enhances Mamba models for segmentation by preventing detail loss via boundary-aware and uncertainty-driven state restoration.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.