VAMamba: An Efficient Visual Adaptive Mamba for Image Restoration

2025-09-27

AI Summary

Existing Mamba-based image inpainting methods suffer from fixed scanning orders and inefficient feature utilization, limiting their adaptability to diverse degradations. To address this, we propose VAMamba, a vision-adaptive framework. First, we introduce the QCLAM caching mechanism, enabling low-rank dynamic fusion of historical features. Second, we design GPS-SS2D—a greedy path-searching strategy that leverages ViT-based importance scoring to generate optimal, degradation-aware scanning paths. Third, we enhance the SS2D architecture with LoRA adaptation, FIFO caching, and low-rank parameter updates to improve modeling flexibility and computational efficiency. Extensive experiments demonstrate that VAMamba consistently outperforms state-of-the-art methods across multiple inpainting benchmarks, achieving superior trade-offs between reconstruction fidelity and inference speed. Ablation studies validate the effectiveness of both adaptive scanning and memory-augmented mechanisms.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recent Mamba-based image restoration methods have achieved promising results but remain limited by fixed scanning patterns and inefficient feature utilization. Conventional Mamba architectures rely on predetermined paths that cannot adapt to diverse degradations, constraining both restoration performance and computational efficiency. To overcome these limitations, we propose VAMamba, a Visual Adaptive Mamba framework with two key innovations. First, QCLAM(Queue-basedCacheLow-rankAdaptiveMemory)enhancesfeaturelearningthrougha FIFO cache that stores historical representations. Similarity between current LoRA-adapted and cached features guides intelligent fusion, enabling dynamic reuse while effectively controlling memorygrowth.Second, GPS-SS2D(GreedyPathScanSS2D)introducesadaptive scanning. A Vision Transformer generates score maps to estimate pixel importance, and a greedy strategy de termines optimal forward and backward scanning paths. These learned trajectories replace rigid patterns, enabling SS2D to perform targeted feature extraction. The integration of QCLAM and GPS-SS2D allows VAMamba to adaptively focus on degraded regions while maintaining high computational efficiency. Extensive experiments across diverse restoration tasks demonstrate that VAMamba consistently outperforms existing approaches in both restoration quality and efficiency, establishing new benchmarks for adaptive image restoration. Our code is available at https://github.com/WaterHQH/VAMamba.

Problem

• Overcoming fixed scanning patterns in Mamba-based image restoration
• Enhancing feature utilization efficiency for diverse degradation types
• Enabling adaptive focus on degraded regions while maintaining computational efficiency

Innovation

• QCLAM enables dynamic feature reuse with FIFO cache
• GPS-SS2D introduces adaptive scanning using greedy strategy
• Integration allows targeted focus on degraded regions efficiently