Skip to content
Book Open access

UVEMamba: Motion-Rectified Spatiotemporal Mamba for Underwater Video Enhancement

Nov 2026 · Proceedings of the 34th ACM International Conference on Multimedia · 0 citations · 41 references

TL;DR

UVEMamba is presented, a framework specifically engineered to bridge the gap between linear-complexity sequence modeling and complex aquatic degradation, and contains a module that explicitly rectifies non-rigid motion offsets before state evolution, ensuring that Mamba's hidden states capture stable trajectories rather than corrupted noise.

Abstract

Underwater video enhancement (UVE) is essential for marine exploration, yet non-rigid refractive distortions and frequency-dependent attenuation often compromise the efficacy of standard restoration models. Although State Space Models (SSMs) like Mamba provide computational efficiency, their traditional direct-scan paradigm fails to accommodate the stochastic feature shifts and turbulent motion inherent in aquatic scenes. In this paper, we present UVEMamba, a framework specifically engineered to bridge the gap between linear-complexity sequence modeling and complex aquatic degradation. To address temporal instability, we propose the Motion-Rectified Spatiotemporal Mamba (MR-STM). By integrating a Sparse Selective Perception Module, this module explicitly rectifies non-rigid motion offsets before state evolution, ensuring that Mamba's hidden states capture stable trajectories rather than corrupted noise. To further decouple complex degradations, we develop the Spatial-Frequency Mamba Block (SFMB), which alternates between spatial geometric correction and wavelet-based frequency aggregation. This isolates global chromatic casts from high-frequency scattering, allowing for targeted artifact suppression and detail preservation. Complementing this architecture, our Unified Spatio-Temporal Color Loss (USTC-Loss) employs motion-aware contrastive learning to enforce long-term consistency, mitigating the common trade-off between structural fidelity and visual perception. Extensive experiments on four datasets demonstrate that our approach generalizes well, delivering high-fidelity and temporally stable restoration for complex underwater videos.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.