SADe improves few-shot segmentation with weak annotation cleaning

Researchers have developed SADe, a novel layer designed to improve few-shot segmentation by cleaning up weak support annotations. This method uses sparse autoencoder atom evidence to estimate the reliability of support patches, effectively removing distractors and background noise. SADe can be integrated with existing few-shot segmentation models without altering their query-side inference, demonstrating significant performance gains across various weak-support scenarios and outperforming other methods on specific prompt-shot combinations. AI

IMPACT Enhances the robustness of segmentation models to imperfect annotations, potentially improving real-world applications.

RANK_REASON This is a research paper detailing a new method for few-shot segmentation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Few-Shot Segmentation
• FSS-1000
• SAM3
• Sparse-Atom Support Decontamination
• Sparse Autoencoder

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →