Rethinking Visual Privacy: A Compositional Privacy Risk Framework for Severity Assessment with VLMs

Efthymios Tsaprazlis [email protected], Tiantian Feng, Anil Ramakrishna, Sai Praneeth Karimireddy, Rahul Gupta, Shrikanth Narayanan · 2026

Read Rethinking Visual Privacy: A Compositional Privacy Risk Framework for Severity Assessment with VLMs by Efthymios Tsaprazlis [email protected], Tiantian Feng, Anil Ramakrishna, Sai Praneeth Karimireddy, Rahul Gupta, Shrikanth Narayanan (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-2603-21573.txt.

Description

Tsaprazlis and co-authors argue that privacy risk in images is compositional, not binary. Harmless details can combine into a serious exposure, yet benchmarks mostly label images private or not. Their Compositional Privacy Risk Taxonomy sorts visual attributes by how identifying they are alone and in combination. It defines four severity levels plus a scoring function for continuous risk. They built a 6,700-image dataset aligned with the taxonomy. Frontier vision models matched the graded scores with guidance, but smaller models struggled.

Catalog record

CREATOREfthymios Tsaprazlis [email protected], Tiantian Feng, Anil Ramakrishna, Sai Praneeth Karimireddy, Rahul Gupta, Shrikanth Narayanan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65768
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.21573

Browse the full library · About the Tower of Babel