RAUM-Net: Regional Attention and Uncertainty-aware Mamba Network

2025-06-27

AI Summary

Fine-grained visual classification (FGVC) faces challenges including subtle inter-class distinctions, scarcity of labeled data, and poor robustness to occlusion. To address these, we propose the first semi-supervised framework for FGVC, innovatively integrating Mamba-based sequence modeling into this domain. Our method incorporates a region-aware attention mechanism to emphasize discriminative local features and employs Bayesian uncertainty estimation to dynamically select high-confidence pseudo-labels. This design significantly enhances generalization under few-shot and occluded conditions. We achieve state-of-the-art performance on multiple benchmarks—including CUB-200, Stanford Cars, and FGVC-Aircraft—particularly excelling under low labeling ratios (≤10%) and severe occlusion, where we substantially outperform existing approaches. The implementation is publicly available.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Semi-Supervised Learning
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• 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
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Fine Grained Visual Categorization (FGVC) remains a challenging task in computer vision due to subtle inter class differences and fragile feature representations. Existing methods struggle in fine grained scenarios, especially when labeled data is scarce. We propose a semi supervised method combining Mamba based feature modeling, region attention, and Bayesian uncertainty. Our approach enhances local to global feature modeling while focusing on key areas during learning. Bayesian inference selects high quality pseudo labels for stability. Experiments show strong performance on FGVC benchmarks with occlusions, demonstrating robustness when labeled data is limited. Code is available at https://github.com/wxqnl/RAUM Net.

Problem

• Addresses Fine Grained Visual Categorization challenges with subtle inter class differences
• Improves feature modeling in semi supervised learning with scarce labeled data
• Enhances robustness to occlusions and limited labeled data in FGVC

Innovation

• Mamba-based feature modeling for FGVC
• Region attention focuses on key areas
• Bayesian uncertainty selects pseudo labels