AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Beyond Walking: A Large-Scale Image-Text Benchmark for Text-based Person Anomaly Search

Shuyu Yang, Yaxiong Wang, Li Zhu, Zhedong Zheng

ICCV 2025 · Vision-language and multimodal › Human and embodied modeling · approach: Multimodal retrieval-augmented modeling · 17 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train X-VLM on 0.1M synthetic PAB pairsX-VLM · PAB testR@1: 71.94 -> 81.95Improved
add pose-aware image encoder (PE) (ablation)X-VLM baseline · PAB test (1M)R@1: 83.47 -> 84.13Improved
add identity-based hard negative mining (IHNM) (ablation)X-VLM baseline · PAB test (1M)R@1: 83.47 -> 84.48Improved
add both PE and IHNM (ablation)X-VLM baseline · PAB test (1M)R@1: 83.47 -> 84.93Improved
add both PE and IHNMX-VLM · PAB test (0.1M)R@1: 81.95 -> 83.06Improved
train CMP on 1M instead of 0.1M pairsCMP · PAB testR@1: 83.06 -> 84.93Improved
train CMP on 1M instead of 0.1M pairsCMP · UCC OODR@1: 54.12 -> 55.23Improved
add PE and IHNM (ablation)X-VLM baseline · PAB multi-weather meanR@1: 64.11 -> 67.12Improved
add IHNM only vs full CMP (ablation)CMP · PAB multi-weather Dark + RainR@1: 40.75 -> 39.03Hurt
use behavior search evaluation instead of identity searchCMP · PAB testR@1: 94.34 -> 84.93Hurt
use CMP instead of RaSaRaSa · UCC OOD (0.1M)R@1: 54.12 -> 54.12No clear effect
use only 10% of PAB training data with hard negative sampling (ablation)CMP · PAB testrecallNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy