Purge-Gate: Backpropagation-Free Test-Time Adaptation for Point Clouds Classification via Token Purging

2025-09-11

AI Summary

To address test-time performance degradation in 3D point cloud classification caused by domain shift, this paper proposes Token Purging (PG), a backward-propagation-free test-time adaptation method that dynamically prunes tokens heavily affected by domain shift prior to ViT attention layers. PG introduces two lightweight variants: PG-SP, which leverages source-domain statistics, and PG-SF, guided by the CLS token—both require no source-model fine-tuning. To our knowledge, PG is the first gradient-free, token-level test-time adaptation approach for 3D vision. On ModelNet40-C, ShapeNet-C, and ScanObjectNN-C, PG-SP achieves an average accuracy gain of 10.3% over the previous best backpropagation-free methods. PG-SF further establishes a new state-of-the-art, accelerating inference by 12.4× and reducing memory consumption by 5.5× compared to existing approaches.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Security and Privacy: Privacy-enhancing technologies
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Test-time adaptation (TTA) is crucial for mitigating performance degradation caused by distribution shifts in 3D point cloud classification. In this work, we introduce Token Purging (PG), a novel backpropagation-free approach that removes tokens highly affected by domain shifts before they reach attention layers. Unlike existing TTA methods, PG operates at the token level, ensuring robust adaptation without iterative updates. We propose two variants: PG-SP, which leverages source statistics, and PG-SF, a fully source-free version relying on CLS-token-driven adaptation. Extensive evaluations on ModelNet40-C, ShapeNet-C, and ScanObjectNN-C demonstrate that PG-SP achieves an average of +10.3% higher accuracy than state-of-the-art backpropagation-free methods, while PG-SF sets new benchmarks for source-free adaptation. Moreover, PG is 12.4 times faster and 5.5 times more memory efficient than our baseline, making it suitable for real-world deployment. Code is available at hyperlink{https://github.com/MosyMosy/Purge-Gate}{https://github.com/MosyMosy/Purge-Gate}

Problem

• Mitigating performance degradation from 3D point cloud distribution shifts
• Removing domain-shift affected tokens before attention layers
• Enabling backpropagation-free test-time adaptation for real-world deployment

Innovation

• Token Purging removes domain-shifted tokens
• Backpropagation-free adaptation via attention layer filtering
• CLS-token-driven source-free variant achieves benchmark performance