AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Self-similarity Driven Scale-invariant Learning for Weakly Supervised Person Search

Benzhi Wang, Yang Yang, Jinlin Wu, Guo-jun Qi, Zhen Lei

ICCV 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Supervised person re-identification · 21 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
add scale-invariant loss with multi-scale exemplar branch to baseline (ablation)weakly supervised person search baseline · PRWmAP: 18.8 -> 26.2Improved
add dynamic multi-label learning to baseline (ablation)weakly supervised person search baseline · PRWmAP: 18.8 -> 20.8Improved
add dynamic multi-label learning on top of baseline plus scale-invariant loss (ablation)SSL · CUHK-SYSUmAP: 86.2 -> 87.4Improved
use multiple scales instead of a single scale for exemplars (ablation)SSL · PRWmAP: 24.3 -> 27.1Improved
use a single fixed scale instead of original crop size (ablation)SSL · PRWmAP: 23.9 -> 24.3No clear effect
filter background of cropped exemplars with binary masks (ablation)SSL · PRWmAP: 27.1 -> 28.6Improved
use dynamic threshold for multi-label pseudo labels instead of a fixed threshold (ablation)SSL · CUHK-SYSUmAP: 84.2 -> 87.4Improved
use dynamic threshold instead of fixed threshold (ablation)SSL · PRWmAP: 29.2 -> 30.7Improved
weakly supervised SSL versus fully supervised methodsSSL · PRWmAP: 30.7 vs 53.3 (COAT)Hurt

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