UV-Attack : Physical-World Adversarial Attacks for Person Detection via Dynamic-NeRF-based UV Mapping

Yanjie Li, Wenxuan Zhang, Kaisheng Liang, Bin Xiao · 2025

Read UV-Attack : Physical-World Adversarial Attacks for Person Detection via Dynamic-NeRF-based UV Mapping by Yanjie Li, Wenxuan Zhang, Kaisheng Liang, Bin Xiao (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-2501-05783.txt.

Description

Yanjie Li, Kaisheng Liang, and Bin Xiao craft clothing textures that fool person detectors in the real world, in a paper published in two thousand twenty-five. Earlier attacks printed patches or edited static three-dimensional models, but people bend and move, and success rates stayed low.

The team's method, UV-Attack, leans on dynamic neural radiance fields, which can model a moving human body. A UV mapping step ties the clothing texture to the body's surface, so the pattern deforms naturally with each pose. The system can render the wearer across many actions and viewpoints. It can even invent new actions by sampling the standard body model's parameter space. The challenge is that the texture hides inside the neural model's parameters, so editing it takes care. The resulting attacks keep working across extensive, unseen movements. The work exposes a robustness gap in detectors deployed for surveillance and safety.

Catalog record

CREATORYanjie Li, Wenxuan Zhang, Kaisheng Liang, Bin Xiao
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-05783

Browse the full library · About the Tower of Babel