Exploring Universal Domain Adaptation with CLIP Models: A Calibration Method

Bin Deng · 2025

Read Exploring Universal Domain Adaptation with CLIP Models: A Calibration Method by Bin Deng (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/pmc-12731723.txt.

Description

Bin Deng wrote a 2025 study on adapting vision-language models to new image tasks when the target classes are partly unknown. Such universal adaptation must sort known classes and flag new ones. Deng tested leading methods built on a big pretrained model. The foundation model lifted plain baselines a lot.

Yet fancy adaptation methods failed to beat that baseline, a surprise. Careful checks pointed to score calibration as the key lever. A simple automatic temperature scaling sharpened detection of out-of-class images. One learned temperature beat older approaches on most benchmark tasks. Deng offers the method and findings as a base for future work.

Catalog record

CREATORBin Deng
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC12731723

Browse the full library · About the Tower of Babel