Lune

CVPR Findings2026

Co-Adaptive Graph Learning Through Coupled Spectral Refinement for 3D Anomaly Detection

Hanvitha Saraswathi Mukkamala, Arun K Pujari

Abstract

Graph-based learning methods such as Laplacian Eigenmaps, Graph Signal Processing, and Graph Neural Networks typically rely on a fixed graph topology to learn node representations. However, in many real-world settings-especially those involving 3D point-cloud data-the spatial and contextual relationships between points evolve as embeddings are refined. Further, in problems like 3D anomaly detection and Bone-Side Estimation, the task often involves comparing two related point clouds to identify subtle geometric deviations.

To address these challenges, we propose a coupled spectral analysis framework that jointly refines graph structure and node embeddings for adaptive 3D learning. Our first method, CoLE-SR, extends Coupled Laplacian Eigenmaps to iteratively align local geometry across graphs, achieving progressive spectral consistency. The second, CoLE-LSA, employs a Graph-BERT-style encoder that implicitly learns topology through attention coefficients, enabling flexible modeling without explicit Laplacian construction. Evaluated on the MVTec 3D-AD benchmark and bone surface datasets, our framework generalizes effectively across domains and consistently outperforms existing methods, demonstrating the strength of co-evolutionary graph learning for robust and accurate 3D anomaly detection.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

Builds onCited byLeft to right by year; the more cited, the higher.

Builds on10

  1. Scaling Out-of-Distribution Detection for Real-World SettingsDan Hendrycks, Steven Basart, Mantas Mazeika, Andy Zou et al.ICML 2022 · 653 citations
  2. Cross-Domain Graph Anomaly Detection via Anomaly-Aware Contrastive AlignmentQizhou Wang, Guansong Pang, Mahsa Salehi, Wray L. Buntine et al.AAAI 2023 · 51 citations
  3. PointAD: Comprehending 3D Anomalies from Points and Pixels for Zero-shot 3D Anomaly DetectionQihang Zhou, Jiangtao Yan, Shibo He, Wenchao Meng et al.NeurIPS 2024 · 49 citations
  4. Laplacian Canonization: A Minimalist Approach to Sign and Basis Invariant Spectral EmbeddingGeorge Ma, Yifei Wang, Yisen WangNeurIPS 2023 · 31 citations
  5. Revisiting Multimodal Fusion for 3D Anomaly Detection from an Architectural PerspectiveKaifang Long, Guoyang Xie, Lianbo Ma, Jiaqi Liu et al.AAAI 2025 · 17 citations

Related papers

  1. Coupled Laplacian Eigenmaps for Locally-Aware 3D Rigid Point Cloud MatchingMatteo Bastico, Etienne Decencière, Laurent Corté, Yannick Tillier et al.CVPR 2024
  2. Graph Smoothing for Enhanced Local Geometry Learning in Point Cloud AnalysisShangbo Yuan, Jie Xu, Ping Hu, Xiaofeng Zhu et al.AAAI 2026
  3. Modeling Spectral Energy Shifts in Spatio-Temporal Graph Anomaly DetectionYilin Liu, Hongchao Zhang, Ahmad Taha, Taylor T Johnson et al.ICML 2026
  4. Graph Spectral Perturbation for 3D Point Cloud Contrastive LearningYuehui Han, Jiaxin Chen, Jianjun Qian, Jin XieACM MM 2023 · 6 citations
  5. 3D Test-Time Adaptation via Graph Spectral Driven Point ShiftXin Wei, Qin Yang, Yijie Fang, Mingrui Zhu et al.ICCV 2025
Dusk over the sea between two cliffs drawn in fine vertical lines