← All publications

Journal · 2026

Video and Scene Understanding

Probing 3D Anomalies via Multi-View Registration and Dual-Residual Analysis

Yuxing Yang, Zeyu Fu, Yang Li, Liewei Wang, Siyue Yu, Jimin Xiao

Neurocomputing

Framework for 3D anomaly detection: multi-view generation, registration pipeline and anomaly scoring.

Abstract

Overview

Due to the sparse, unordered, and low information density characteristics of point cloud data, 3D anomaly detection is a challenging task, especially under the conditions of few and only normal training data. Recently, registration-based point cloud anomaly detection has gained increasing popularity. To address the inherent local optimum and pose ambiguity issues of registration-based methods, we revisit registration from a multi-view perspective. Then, a downsampled coarse-to-fine registration pipeline is utilized to align the test samples with the enhanced normal references. Geometric and feature residuals are combined to capture both point-wise spatial deviations and differences in the high-dimensional feature space distribution. Our framework is evaluated on public Anomaly-ShapeNet and Real3D-AD datasets, along with other state-of-the-art work at the object and point levels. A detailed ablation study is conducted to analyze the effect of multi-view generation strategy, dual-branch residuals, and input features. Extensive experiments validate the effectiveness of our proposed method, achieving improvements of 4.1% and 1.8% at the object-level AUROC, respectively. The code will be released to support further research.

Citation

BibTeX

@article{Yang_2026, title={Probing 3D anomalies via multi-view registration and dual-residual analysis}, volume={684}, ISSN={0925-2312}, url={http://dx.doi.org/10.1016/j.neucom.2026.133558}, DOI={10.1016/j.neucom.2026.133558}, journal={Neurocomputing}, publisher={Elsevier BV}, author={Yang, Yuxing and Fu, Zeyu and Li, Yang and Wang, Liewei and Yu, Siyue and Xiao, Jimin}, year={2026}, month=July, pages={133558} }