Self-Supervised and Generalizable Tokenization for CLIP-Based 3D Understanding

2025-05-24

AI Summary

Existing CLIP-based 3D scene understanding methods rely on k-NN or radius-based tokenization, making them susceptible to dataset-specific spatial scale biases and limiting generalization. This paper proposes S4Token—the first unsupervised, scale-invariant 3D tokenization framework tailored for CLIP—integrating superpoint grouping, coordinate-scale normalization, and semantic-aware feature propagation. We introduce a novel joint training paradigm comprising masked point modeling, contrastive clustering loss, and multi-view image–3D cross-modal distillation, enabling self-supervised representation learning with a frozen CLIP backbone. Evaluated on ScanNet and S3DIS benchmarks, S4Token achieves +12.6% average accuracy gains in zero-shot 3D retrieval and open-vocabulary segmentation over conventional tokenization approaches, demonstrating superior cross-domain generalization capability.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Unsupervised & Self-Supervised Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Vision-language models like CLIP can offer a promising foundation for 3D scene understanding when extended with 3D tokenizers. However, standard approaches, such as k-nearest neighbor or radius-based tokenization, struggle with cross-domain generalization due to sensitivity to dataset-specific spatial scales. We present a universal 3D tokenizer designed for scale-invariant representation learning with a frozen CLIP backbone. We show that combining superpoint-based grouping with coordinate scale normalization consistently outperforms conventional methods through extensive experimental analysis. Specifically, we introduce S4Token, a tokenization pipeline that produces semantically-informed tokens regardless of scene scale. Our tokenizer is trained without annotations using masked point modeling and clustering-based objectives, along with cross-modal distillation to align 3D tokens with 2D multi-view image features. For dense prediction tasks, we propose a superpoint-level feature propagation module to recover point-level detail from sparse tokens.

Problem

• Improving cross-domain generalization in 3D tokenization for CLIP
• Achieving scale-invariant 3D representation learning with frozen CLIP
• Enabling semantic 3D tokenization without manual annotations

Innovation

• Scale-invariant 3D tokenizer with CLIP
• Superpoint grouping and scale normalization
• Self-supervised training with cross-modal distillation