PaIR: Partition-Based Information Rebalancing for Robust Text-Based Person Search

Luda Wang, Jiabao Li, Xinpan Yuan, Ningdan Zhang · 2026

Read PaIR: Partition-Based Information Rebalancing for Robust Text-Based Person Search by Luda Wang, Jiabao Li, Xinpan Yuan, Ningdan Zhang (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/pmc-13608582.txt.

Description

Luda Wang and three co-authors present PaIR, a framework for text-based person search, in work published in 2026. The task matches a written description to the right pedestrian image, but images are rich and messy while texts are sparse and sometimes wrong, so the two sides carry uneven information. Their design builds global and part-level views of both sides on a shared dual encoder.

Noise filters strip background clutter from images and weak tokens from text. A part alignment module splits the body into semantic parts, soft-matches them and fills gaps where the text omits a part. Joint global-local training then ties the levels together, and tests on three benchmarks improved several baselines.

Catalog record

CREATORLuda Wang, Jiabao Li, Xinpan Yuan, Ningdan Zhang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13608582

Browse the full library · About the Tower of Babel