Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards High-resolution 3D Anomaly Detection via Group-Level Feature Contrastive Learning

Hongze Zhu, Guoyang Xie, Chengbin Hou, Tao Dai, Can Gao, Jinbao Wang, Linlin Shen

TL;DR

High-resolution 3D anomaly detection (HRPCD-AD) faces challenges from massive point counts, anisotropic transformer-based embeddings, and sparsity of anomalies. The authors propose Group3AD, a group-level feature contrastive learning framework that combines Intercluster Uniformity Network (IUN), Intracluster Alignment Network (IAN), and Adaptive Group-Center Selection (AGCS) to achieve uniform and tightly clustered group features while focusing inference on likely anomalous regions. The method, built on a pre-trained encoder and memory-bank scoring, demonstrates state-of-the-art performance on Real3D-AD with notable gains over Reg3D-AD in object- and point-level AUROC/AUPR, and is validated through extensive ablations of IUN, IAN, and AGCS. Overall, Group3AD offers a scalable, resolution-agnostic approach that effectively exploits HR spatial information for accurate and efficient HRPCD anomaly localization in industrial settings.

Abstract

High-resolution point clouds~(HRPCD) anomaly detection~(AD) plays a critical role in precision machining and high-end equipment manufacturing. Despite considerable 3D-AD methods that have been proposed recently, they still cannot meet the requirements of the HRPCD-AD task. There are several challenges: i) It is difficult to directly capture HRPCD information due to large amounts of points at the sample level; ii) The advanced transformer-based methods usually obtain anisotropic features, leading to degradation of the representation; iii) The proportion of abnormal areas is very small, which makes it difficult to characterize. To address these challenges, we propose a novel group-level feature-based network, called Group3AD, which has a significantly efficient representation ability. First, we design an Intercluster Uniformity Network~(IUN) to present the mapping of different groups in the feature space as several clusters, and obtain a more uniform distribution between clusters representing different parts of the point clouds in the feature space. Then, an Intracluster Alignment Network~(IAN) is designed to encourage groups within the cluster to be distributed tightly in the feature space. In addition, we propose an Adaptive Group-Center Selection~(AGCS) based on geometric information to improve the pixel density of potential anomalous regions during inference. The experimental results verify the effectiveness of our proposed Group3AD, which surpasses Reg3D-AD by the margin of 5\% in terms of object-level AUROC on Real3D-AD. We provide the code and supplementary information on our website: https://github.com/M-3LAB/Group3AD.

Towards High-resolution 3D Anomaly Detection via Group-Level Feature Contrastive Learning

TL;DR

High-resolution 3D anomaly detection (HRPCD-AD) faces challenges from massive point counts, anisotropic transformer-based embeddings, and sparsity of anomalies. The authors propose Group3AD, a group-level feature contrastive learning framework that combines Intercluster Uniformity Network (IUN), Intracluster Alignment Network (IAN), and Adaptive Group-Center Selection (AGCS) to achieve uniform and tightly clustered group features while focusing inference on likely anomalous regions. The method, built on a pre-trained encoder and memory-bank scoring, demonstrates state-of-the-art performance on Real3D-AD with notable gains over Reg3D-AD in object- and point-level AUROC/AUPR, and is validated through extensive ablations of IUN, IAN, and AGCS. Overall, Group3AD offers a scalable, resolution-agnostic approach that effectively exploits HR spatial information for accurate and efficient HRPCD anomaly localization in industrial settings.

Abstract

High-resolution point clouds~(HRPCD) anomaly detection~(AD) plays a critical role in precision machining and high-end equipment manufacturing. Despite considerable 3D-AD methods that have been proposed recently, they still cannot meet the requirements of the HRPCD-AD task. There are several challenges: i) It is difficult to directly capture HRPCD information due to large amounts of points at the sample level; ii) The advanced transformer-based methods usually obtain anisotropic features, leading to degradation of the representation; iii) The proportion of abnormal areas is very small, which makes it difficult to characterize. To address these challenges, we propose a novel group-level feature-based network, called Group3AD, which has a significantly efficient representation ability. First, we design an Intercluster Uniformity Network~(IUN) to present the mapping of different groups in the feature space as several clusters, and obtain a more uniform distribution between clusters representing different parts of the point clouds in the feature space. Then, an Intracluster Alignment Network~(IAN) is designed to encourage groups within the cluster to be distributed tightly in the feature space. In addition, we propose an Adaptive Group-Center Selection~(AGCS) based on geometric information to improve the pixel density of potential anomalous regions during inference. The experimental results verify the effectiveness of our proposed Group3AD, which surpasses Reg3D-AD by the margin of 5\% in terms of object-level AUROC on Real3D-AD. We provide the code and supplementary information on our website: https://github.com/M-3LAB/Group3AD.
Paper Structure (29 sections, 16 equations, 5 figures, 9 tables, 2 algorithms)

This paper contains 29 sections, 16 equations, 5 figures, 9 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. RELATED WORK
  3. 2D Anomaly Detection
  4. RGB-D Anomaly Detection
  5. High-resolution 3D Anomaly Detection
  6. APPROACH
  7. Problem Definition
  8. Group3AD
  9. Intercluster Uniformity Network (IUN)
  10. Intracluster Alignment Network (IAN)
  11. Adaptive Group-Center Selection (AGCS)
  12. Overall Pipeline
  13. EXPERIMENTS
  14. Experimental Settings
  15. Results and Analysis
  16. ...and 14 more sections

Figures (5)

  • Figure 1: Ideal feature distribution of normal point clouds and abnormalies for the high-resolution 3D-AD task. Group-level features are used to express structural information.
  • Figure 2: The pipeline of Group3AD, which consists of three main components. (1) Group-Level Feature Extraction extracts group-level features from the input 3D point clouds. (2) Intercluster Uniformity Network (IUN) and Intracluster Alignment Network (IAN) enhance the feature separation between clusters and tighten the distribution within clusters, respectively, for improving anomaly detection accuracy. (3) Adaptive Group-Center Selection (AGCS), used during inference, dynamically focuses on regions with potential anomalies by adjusting the sampling density based on geometric information. This structured approach ensures efficient anomaly detection in complex 3D environments.
  • Figure 3: Flowchat of group-level feature distribution, constrained by IUN and IAN. The basic idea is to minimize the intra-group distance and maximize the inter-group distance.
  • Figure 4: Adaptive group-center selection (AGCS). AGCS adaptively selects the points most likely to be within the anomaly region in the selection of group centers via FPFH features.
  • Figure 5: Visualization results obtained by Group3AD. Different colors indicate different groups selected by AGCS. The red circle represents anomaly area.