AI Paper Atlas

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

Graph Sampling Based Deep Metric Learning for Generalizable Person Re-Identification

Shengcai Liao, Ling Shao

CVPR 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Supervised person re-identification · 129 citations (OpenAlex)

Check 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
Replace random PK sampling with nearest-neighbor graph samplingQAConv · Cross-dataset person re-identification, including MSMT17 training → CUHK03-NPRank-1: 18.6 -> 20.9Improved
Replace random PK sampling with nearest-neighbor graph samplingQAConv · RandPerson training → MSMT17Rank-1: 40.3 -> 45.1Improved
Use graph sampling instead of PK samplingOSNet-IBN · MSMT17 (all) training → Market-1501Rank-1: 67.9 -> 71.3Improved
Use graph sampling instead of PK samplingTransMatcher · Market-1501 training → CUHK03-NPRank-1: 22.9 -> 22.2Hurt
Replace hard triplet loss with binary cross-entropy loss (ablation)QAConv-GS · Market-1501 training → CUHK03-NPRank-1: 19.1 -> 16.4Hurt
Use 256×128 instead of 384×128 input (ablation)QAConv-GS · Market-1501 training → CUHK03-NPRank-1: 19.1 -> 16.9Hurt
Use class centers for graph construction instead of one sampled example per class (ablation)QAConv-GS · Market-1501 training → CUHK03-NPRank-1: 19.1 -> 15.4Hurt
Use class centers for graph construction instead of one sampled example per class (ablation)QAConv-GS · MSMT17 training → CUHK03-NPRank-1: 20.9 -> 15.3Hurt
Apply gradient clipping (ablation)QAConv-GS · MSMT17 traininggeneralization performanceNo clear effect
Apply gradient clipping (ablation)QAConv-GS · Market-1501 and RandPerson traininggeneralization performanceImproved
Increase triplet-loss margin above 32 (ablation)QAConv-GS · MSMT17 trainingmAccHurt
Increase the number of images per class beyond K=2 with hard triplet loss (ablation)QAConv-GS · Training convergenceconvergenceHurt

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