AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

One-Shot Knowledge Transfer for Scalable Person Re-Identification

Longhua Li, Lei Qi, Xin Geng

ICCV 2025 · Label-efficient and robust learning › Knowledge distillation · approach: Few-shot and metric distillation · 2 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
use trained teacher weights instead of a randomly initialized teacher (ablation)OSKT Res-50-S1 · Market1501 single scenario (ablation)mAP: 54.7 -> 75.7Improved
use trained teacher weights instead of a randomly initialized teacher (ablation)OSKT ViT-S-S1 · Market1501 single scenario (ablation)mAP: 33.0 -> 74.2Improved
initialize the weight chain with cluster centers instead of random clustering (ablation)OSKT Res-50-S1 · Market1501 single scenario (ablation)mAP: 55.1 -> 75.7Improved
initialize with clustering instead of inverse clustering (ablation)OSKT Res-50-S1 · Market1501 single scenario (ablation)mAP: 46.1 -> 75.7Improved
use Euclidean distance for ResNet and cosine for ViT rather than swapped (ablation)OSKT ViT-S-S1 · Market1501 single scenario (ablation)mAP: 64.9 -> 74.2Improved
use Euclidean distance for ResNet rather than cosine (ablation)OSKT Res-50-S1 · Market1501 single scenario (ablation)mAP: 74.2 -> 75.7No clear effect
add progressive refinement of the weight chain (ablation)OSKT Res-50-S1 · Market1501 single scenario (ablation)mAP: 52.2 -> 75.7Improved
add progressive refinement of the weight chain (ablation)OSKT ViT-S-S1 · Market1501 single scenario (ablation)mAP: 56.6 -> 74.2Improved
train teacher loss L_T jointly during refinement (ablation)OSKT Res-50-S1 · Market1501 single scenario (ablation)mAP: 62.7 -> 75.7Improved
use the default alpha schedule for L_ref instead of swapped (ablation)OSKT Res-50-S1 · Market1501 single scenario (ablation)mAP: 75.6 -> 75.7No clear effect
OSKT instead of DepGraph pruning for Res-50-S1student generation · Market1501 single scenario (ResNet50 teacher)mAP: 61.3 -> 75.7Improved
OSKT initialization instead of ImageNet-free pretrain on OSNet width 0.25OSNet 0.25 · CUHK03 (pretrained on MSMT17)mAP: 58.0 -> 60.6Improved

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