AdapTS: Lightweight Teacher-Student Approach for Multi-Class and Continual Visual Anomaly Detection

Manuel Barusco, Davide Dalle Pezze, Francesco Borsatti, Gian Antonio Susto · 2026

Read AdapTS: Lightweight Teacher-Student Approach for Multi-Class and Continual Visual Anomaly Detection by Manuel Barusco, Davide Dalle Pezze, Francesco Borsatti, Gian Antonio Susto (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-2603-17530.txt.

Description

Barusco, Dalle Pezze, Borsatti, and Susto present AdapTS for visual anomaly detection on factory lines. Most methods handle one product class and never learn new ones over time. AdapTS shares one frozen backbone and adds small trainable adapters for the student path. A prototype step picks the right adapters at run time with 99 percent accuracy, they report. The lightest variant adds only 8 megabytes of memory.

Catalog record

CREATORManuel Barusco, Davide Dalle Pezze, Francesco Borsatti, Gian Antonio Susto
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34157
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.17530

Browse the full library · About the Tower of Babel