Overcoming Support Dilution for Robust Few-shot Semantic Segmentation

2025-01-23

AI Summary

To address the “support dilution” problem in few-shot semantic segmentation—where enlarging the support set introduces redundant or low-quality samples that degrade generalization—this paper proposes a contribution-driven support set optimization framework. Methodologically, it (1) designs a learnable contribution metric to quantify each support sample’s discriminative utility for query segmentation; (2) introduces a Symmetric Correlation (SC) module to model bidirectional contextual dependencies in feature space; and (3) integrates an image-level dynamic pruning mechanism to adaptively select high-contribution samples. Embedded within a meta-learning pipeline, the framework enables end-to-end optimization. Experiments on PASCAL-5i and COCO-20i demonstrate significant improvements over state-of-the-art methods, validating robustness to distribution shifts and real-world scenarios. This work is the first to systematically formulate and mitigate the support dilution effect, establishing a principled approach to support set quality control in few-shot segmentation.

Technology Category

• Computer Vision: Segmentation
• Search and Optimization: Learning to Search
• Machine Learning: Semi-Supervised Learning

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• 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: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Few-shot Semantic Segmentation (FSS) is a challenging task that utilizes limited support images to segment associated unseen objects in query images. However, recent FSS methods are observed to perform worse, when enlarging the number of shots. As the support set enlarges, existing FSS networks struggle to concentrate on the high-contributed supports and could easily be overwhelmed by the low-contributed supports that could severely impair the mask predictions. In this work, we study this challenging issue, called support dilution, our goal is to recognize, select, preserve, and enhance those high-contributed supports in the raw support pool. Technically, our method contains three novel parts. First, we propose a contribution index, to quantitatively estimate if a high-contributed support dilutes. Second, we develop the Symmetric Correlation (SC) module to preserve and enhance the high-contributed support features, minimizing the distraction by the low-contributed features. Third, we design the Support Image Pruning operation, to retrieve a compact and high quality subset by discarding low-contributed supports. We conduct extensive experiments on two FSS benchmarks, COCO-20i and PASCAL-5i, the segmentation results demonstrate the compelling performance of our solution over state-of-the-art FSS approaches. Besides, we apply our solution for online segmentation and real-world segmentation, convincing segmentation results showing the practical ability of our work for real-world demonstrations.

Problem

• Few-shot Semantic Segmentation
• Support Distribution Issue
• Stability and Reliability

Innovation

• Quality Assessment Criterion
• Protection and Enhancement Mechanism
• Filtering Operation for Sample Quality