Multi-class normality

Project information

AVSS 2026 Paper titled:

"Diffusion-Based Multi-Class Normality for OOD Detection: An Application to CDP Authentication"

In this work, we propose a multi-class normality approach based on diffusion models for out-of-distribution (OOD) detection and apply it to copy detection pattern (CDP) authentication. We consider authentic copy detection patterns as in-distribution classes and counterfeit patterns as out-of-distribution samples, and compute a reconstruction-based signal by masking regions of the input and comparing them with their corresponding reconstructions generated by a conditioned diffusion model.