Online Gaussian Test-Time Adaptation of Vision-Language Models

Clément Fuchs, Maxime Zanella *, Christophe De Vleeschouwer · 2025

Read Online Gaussian Test-Time Adaptation of Vision-Language Models by Clément Fuchs, Maxime Zanella *, Christophe De Vleeschouwer (2025) 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-2501-04352.txt.

Description

Clement Fuchs, Maxime Zanella, and Christophe De Vleeschouwer adapt vision language models on the fly with a Gaussian method in a 2025 paper. Test time adaptation tunes a model on incoming data to handle new conditions. Past methods lean on settings fitted per data set, which limits use on unseen tasks.

Their method, OGA, models image features with Gaussian curves and folds in the model's built in zero shot knowledge through a clear statistical frame. Settings stay fixed across all data sets. It beat prior methods on most sets and gained more when joined with few shot tricks. The authors also argue test runs vary too much for three run averages, and propose a stricter score.

Catalog record

CREATORClément Fuchs, Maxime Zanella *, Christophe De Vleeschouwer
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-04352

Browse the full library · About the Tower of Babel