mlsanityData to 14 Oct 2026

White Aggregation and Restoration for Few-shot 3D Point Cloud Semantic Segmentation

arXiv 2509.13907 ↗First posted 17 Sep 2025cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Few-shot 3D Point Cloud Semantic Segmentation (FS-PCS) aims to predict per-point labels for an unlabeled point cloud, given only a few labeled examples. To extract representations from the limited labeled set, existing methods have constructed prototypes with Farthest Point Sampling (FPS). However, we found that this convention results in performance instability due to its sensitivity to FPS-induced variations, while the prototype generation process remains underexplored in the field. This motivates us to investigate deterministic prototype generation method based on attention mechanism. Despite its potential, we found that vanilla attention module suffers from the distributional gap between prototypical tokens and support features. To overcome this, we provide a simple approach, White Aggregation and Restoration Module (WARM), which resolves the misalignment by wrapping cross-attention with whitening and coloring transformations. Specifically, whitening aligns the features to tokens before the attention process, and coloring subsequently restores the original distribution to the attended tokens. This design enables robust attention, thereby generating prototypes that capture the semantic relationships in support features. WARM achieves state-of-the-art performance with a significant margin on the S3DIS dataset, and competitive performance on the ScanNet dataset. Further experiments demonstrate its effectiveness in deterministic prototype generation. Code is publicly available at: https://github.com/JiyunIm00/WARM.git

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 37 of its 43 references that are in the map. First posted on arXiv 17 Sep 2025.

Builds on

37 of 43 references.

References · 37

  1. 2017Attention Is All You Needcited by 60,361
  2. 2020End-to-End Object Detection with Transformerscited by 7,414
  3. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,117
  4. 2017Prototypical Networks for Few-shot Learningcited by 3,015
  5. 2021Masked-attention Mask Transformer for Universal Image Segmentationcited by 2,351
  6. 2016Matching Networks for One Shot Learningcited by 2,255
  7. 2022Language-driven Semantic Segmentationcited by 717
  8. 2021DETR3D: 3D Object Detection from Multi-view Images via 3D-to-2D Queriescited by 658
  9. 2021Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depthcited by 419
  10. 2018How Does Batch Normalization Help Optimization?cited by 357
  11. 2021An End-to-End Transformer Model for 3D Object Detectioncited by 329
  12. 2022Mask3D: Mask Transformer for 3D Semantic Instance Segmentationcited by 303
25 more
  1. 2019Revealing the Dark Secrets of BERTcited by 301
  2. 2022Stratified Transformer for 3D Point Cloud Segmentationcited by 167
  3. 2020Part-aware Prototype Network for Few-shot Semantic Segmentationcited by 162
  4. 2023Stabilizing Transformer Training by Preventing Attention Entropy Collapsecited by 161
  5. 2022Masked Discrimination for Self-Supervised Learning on Point Cloudscited by 155
  6. 2019Modeling Point Clouds with Self-Attention and Gumbel Subset Samplingcited by 151
  7. 2022Self-Support Few-Shot Semantic Segmentationcited by 96
  8. 2018Decorrelated Batch Normalizationcited by 83
  9. 2020Few-shot 3D Point Cloud Semantic Segmentationcited by 77
  10. 2019SampleNet: Differentiable Point Cloud Samplingcited by 74
  11. 2022Feature-Proxy Transformer for Few-Shot Segmentationcited by 58
  12. 2022Unsupervised Point Cloud Representation Learning with Deep Neural Networks: A Surveycited by 54
  13. 2024Dynamic Adapter Meets Prompt Tuning: Parameter-Efficient Transfer Learning for Point Cloud Analysiscited by 53
  14. 2023Prototype Adaption and Projection for Few- and Zero-shot 3D Point Cloud Semantic Segmentationcited by 26
  15. 2024Multimodality Helps Few-shot 3D Point Cloud Semantic Segmentationcited by 20
  16. 2023Boosting Few-shot 3D Point Cloud Segmentation via Query-Guided Enhancementcited by 19
  17. 2024Rethinking Few-shot 3D Point Cloud Semantic Segmentationcited by 18
  18. 2024No Time to Train: Empowering Non-Parametric Networks for Few-shot 3D Scene Segmentationcited by 18
  19. 2023Pixel-Aligned Recurrent Queries for Multi-View 3D Object Detectioncited by 14
  20. 2025Temporal Alignment-Free Video Matching for Few-shot Action Recognitioncited by 10
  21. 2023Threshold-Consistent Margin Loss for Open-World Deep Metric Learningcited by 7
  22. 2024Correlations Are Ruining Your Gradient Descentcited by 7
  23. 2024Localization and Expansion: A Decoupled Framework for Point Cloud Few-shot Semantic Segmentationcited by 6
  24. 2024Deep Learning on 3D Semantic Segmentation: A Detailed Reviewcited by 6
  25. 2024Activating Self-Attention for Multi-Scene Absolute Pose Regressioncited by 2

Built on by

Not cited yet.