Two Steps Are All You Need: Efficient 3D Point Cloud Anomaly Detection with Consistency Models
Pranav A, Shashank B, Pranav Siddappa, Dominik Seuss, Minal Moharir, Subramanya KN · 2026
Read Two Steps Are All You Need: Efficient 3D Point Cloud Anomaly Detection with Consistency Models by Pranav A, Shashank B, Pranav Siddappa, Dominik Seuss, Minal Moharir, Subramanya KN (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-2605-05372.txt.
Description
Factories increasingly scan parts as 3D point clouds and need software to spot defects fast. Pranav A and five co-authors adapt diffusion models to that job, then remove their main drawback. Ordinary diffusion needs many slow denoising steps before it can rebuild a clean shape for comparison.
Their consistency approach predicts the clean geometry in one or two steps instead. A new hybrid loss pushes the rebuild toward defect-free data. The system ran up to eighty times faster than the leading earlier method, with no graphics card needed, while keeping strong detection scores on two standard test sets.
Catalog record
| CREATOR | Pranav A, Shashank B, Pranav Siddappa, Dominik Seuss, Minal Moharir, Subramanya KN |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.05372