Visual Mamba enhances low-light and underwater videos with state-space models

Researchers have developed BVI-Mamba, a novel framework for enhancing videos captured in low-light and underwater conditions. This new method utilizes a Visual State Space (VSS) model to reduce computational demands and memory usage compared to existing AI-based tools. The framework includes a feature alignment module for registering frame displacements and an enhancement module employing VSS blocks for noise removal and brightness adjustment. Experiments indicate that BVI-Mamba surpasses Transformer and convolution-based models in video enhancement tasks. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Offers a more computationally efficient approach to video enhancement, potentially improving performance in specialized environments.

RANK_REASON Academic paper introducing a new model architecture for video enhancement.

Read on arXiv cs.CV →

• BVI-Mamba
• Visual State Space
• UNet
• Nantheera Anantrasirichai
• arXiv
• Hugging Face

• paper
• other