A Semantically Disentangled Unified Model for Multi-category 3D Anomaly Detection

2026-03-26

AI Summary

This work addresses the challenges of semantic prior bias and unreliable anomaly scoring in multi-category 3D anomaly detection, which arise from entangled cross-category features. To mitigate these issues, the authors propose a unified semantic-disentangled model that performs anomaly detection and localization through conditional feature reconstruction. The approach integrates three key components: coarse-to-fine global tokenization, category-conditional contrastive learning, and a geometry-guided decoder, collectively alleviating feature entanglement and enhancing generalization and reliability across diverse object categories. Experimental results demonstrate significant performance gains, with object-level AUROC improvements of 2.8% on Real3D-AD and 9.1% on Anomaly-ShapeNet, achieving state-of-the-art results among both unified and category-specific models.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

3D anomaly detection targets the detection and localization of defects in 3D point clouds trained solely on normal data. While a unified model improves scalability by learning across multiple categories, it often suffers from Inter-Category Entanglement (ICE)-where latent features from different categories overlap, causing the model to adopt incorrect semantic priors during reconstruction and ultimately yielding unreliable anomaly scores. To address this issue, we propose the Semantically Disentangled Unified Model for 3D Anomaly Detection, which reconstructs features conditioned on disentangled semantic representations. Our framework consists of three key components: (i) Coarse-to-Fine Global Tokenization for forming instance-level semantic identity, (ii) Category-Conditioned Contrastive Learning for disentangling category semantics, and (iii) a Geometry-Guided Decoder for semantically consistent reconstruction. Extensive experiments on Real3D-AD and Anomaly-ShapeNet demonstrate that our method achieves state-of-the-art for both unified and category-specific models, improving object-level AUROC by 2.8% and 9.1%, respectively, while enhancing the reliability of unified 3D anomaly detection.

Problem

• 3D anomaly detection
• multi-category
• Inter-Category Entanglement
• unified model
• semantic disentanglement

Innovation

• Semantic Disentanglement
• Unified 3D Anomaly Detection
• Inter-Category Entanglement
• Category-Conditioned Contrastive Learning
• Geometry-Guided Reconstruction