BVI-Mamba: Video Enhancement Using a Visual State-Space Model for Low-Light and Underwater Environments

2026-04-26

AI Summary

This work addresses the degradation of visibility and performance in downstream vision tasks caused by noise, low contrast, color imbalance, and blur commonly found in low-light and underwater videos. To tackle these challenges, the authors propose Visual Mamba, a novel framework that introduces visual state space models (VSS) into video enhancement for the first time. Built upon a lightweight UNet-style architecture, Visual Mamba replaces conventional convolutional layers with VSS blocks to enable precise spatiotemporal feature alignment and effective image quality enhancement. The method achieves superior performance over existing Transformer- and convolution-based approaches on both low-light and underwater video enhancement tasks, while significantly reducing computational and memory overhead.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Intelligent Robots: State Estimation

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Videos captured in low-light and underwater conditions often suffer from distortions such as noise, low contrast, color imbalance, and blur. These issues not only limit visibility but also degrade automatic tasks like detection. Post-processing is typically required but can be time-consuming. AI-based tools for video enhancement also demand significantly more computational resources compared to image-based methods. This paper introduces a novel framework, Visual Mamba, designed to reduce memory usage and computational time by leveraging the Visual State Space (VSS) model. The framework consists of two modules: (i) a feature alignment module, where spatio-temporal displacement between input frames is registered in the feature space, and (ii) an enhancement module, where noise removal and brightness adjustment are performed using a UNet-like architecture, with all convolutional layers replaced by VSS blocks. Experimental results show that the Visual Mamba technique outperforms Transformer and convolution-based models in both low-light and underwater video enhancement tasks. Code is available on line at https://github.com/russellllaputa/BVI-Mamba.

Problem

• low-light video
• underwater video
• video enhancement
• image distortion
• computational efficiency

Innovation

• Visual State-Space Model
• Video Enhancement
• Low-Light Imaging
• Underwater Vision
• Efficient Neural Architecture