Cross-Resolution Semantic Transfer for Robust Text-to-Image Retrieval in Low-Resolution Surveillance 6271 CCS: Computing methodologies Visual content-based indexing and retrieval

Wenjie Qian a , Bin Yang b , Xiao Wang a,c , Wenke Huang b , Ling Mei a,c , Xin Xu a,c , Mang Ye b · 2026

Read Cross-Resolution Semantic Transfer for Robust Text-to-Image Retrieval in Low-Resolution Surveillance 6271 CCS: Computing methodologies Visual content-based indexing and retrieval by Wenjie Qian a , Bin Yang b , Xiao Wang a,c , Wenke Huang b , Ling Mei a,c , Xin Xu a,c , Mang Ye b (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-2606-30458.txt.

Description

Wenjie Qian and six colleagues improve finding people in surveillance photos from text descriptions when image quality varies wildly. Text-to-image person search matches a written description to gallery photos. Real cameras produce mixed sharpness in one gallery. Low resolution destroys fine details the text mentions. Mixed galleries also warp the similarity rankings themselves. The authors name these two failure modes explicitly.

Evidence Reliability Collapse covers the unreliable degraded visual tokens. Ranking Distribution Drift covers the distorted neighborhoods across resolutions. Their framework, CRST, answers with three modules. Resolution-conditioned reasoning estimates which tokens can be trusted. Text-guided refinement restores discriminative cues from the description. A transfer step borrows sharp-image neighborhood structure to steady low-resolution ranks. Three standard benchmarks test the approach. Ultra-low-resolution accuracy rises by about five to six points on average. Sharp-image performance is preserved at the same time.

Catalog record

CREATORWenjie Qian a , Bin Yang b , Xiao Wang a,c , Wenke Huang b , Ling Mei a,c , Xin Xu a,c , Mang Ye b
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.30458

Browse the full library · About the Tower of Babel