Anomaly Factory 3D: A Modular Framework for Diverse Pseudo-Anomaly Synthesis in Unsupervised 3D Anomaly Detection

Ali Balapour, Faraz Hach · 2026

Read Anomaly Factory 3D: A Modular Framework for Diverse Pseudo-Anomaly Synthesis in Unsupervised 3D Anomaly Detection by Ali Balapour, Faraz Hach (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-29181.txt.

Description

Balapour and Hach present Anomaly Factory 3D, a tool that fabricates fake defects to train 3D anomaly detectors. Detecting defects in point clouds, sets of points mapping an object's surface, is hard because abnormal samples are scarce and varied, while training data is usually all normal. Detectors that learn from synthetic pseudo-anomalies need a rich supply of them.

Their framework makes diverse fake defects from normal point clouds. A deformation model, defined in local frames from principal component analysis, warps the surface with controlled falloff, direction gating and separate normal and sideways shifts. That yields a broad set of defect presets, from dents to bumps. The authors plug the tool into two detectors, one predicting offsets and one based on reconstruction, showing it transfers across detection styles. On two benchmarks it improved object-level and point-level detection and localisation, with ablations confirming the value of the preset groups.

Catalog record

CREATORAli Balapour, Faraz Hach
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.29181

Browse the full library · About the Tower of Babel