Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Jun 20, 2026

Artificial Intelligence-Based System for Detecting Attention Levels in Students

Artificial Intelligence-Based System for Detecting Attention Levels in Students

Published on: December 15, 2023

Dynamic adaptive multi-view contrastive learning for unsupervised person re-identification.

Zhi-Hua Li¹, Xue-Yan Wang¹, Si-Bao Chen¹

¹MOE Key Laboratory of ICSP, IMIS Laboratory of Anhui Province, Anhui Provincial Key Laboratory of Multimodal Cognitive Computation, and Zenmorn-AHU AI Joint Laboratory, School of Computer Science and Technology, Anhui University, Hefei, 230601, China.

Neural Networks : the Official Journal of the International Neural Network Society

|May 19, 2026

Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Long-Term Kidney Outcomes Following Dialysis-Treated Childhood Acute Kidney Injury: A Population-Based Cohort Study.

Journal of the American Society of Nephrology : JASN·2021

Same author

Synergistic regulation of methylation and SP1 on MAGE-D4 transcription in glioma.

American journal of translational research·2021

Same author

Incidence of Major Adverse Cardiovascular Events and Cardiac Mortality in High-Risk Kidney-Only and Simultaneous Pancreas-Kidney Transplant Recipients.

Kidney international reports·2021

Same author

The laterodorsal tegmentum-ventral tegmental area circuit controls depression-like behaviors by activating ErbB4 in DA neurons.

Molecular psychiatry·2021

Same author

Frequency splicing code-based Brillouin optical time domain collider for fast dynamic measurement.

Optics express·2021

Same author

Michelson interferometer based phase demodulation for stable time transfer over 1556 km fiber links.

Optics express·2021

This summary is machine-generated.

This study introduces Dynamic Adaptive Multi-view Contrastive Learning (DAMCL) to improve unsupervised person re-identification (Re-ID). DAMCL effectively reduces noise from camera variations and enhances learning accuracy using dynamic proxy and distillation modules.

Area of Science:

Computer Vision
Machine Learning
Artificial Intelligence

Background:

Unsupervised person re-identification (Re-ID) commonly uses clustering for pseudo-label generation in contrastive learning.
Camera variations introduce noise into pseudo-labels, hindering contrastive learning performance due to inaccurate proxy construction and sensitivity to noisy labels.

Purpose of the Study:

To propose a novel framework, Dynamic Adaptive Multi-view Contrastive Learning (DAMCL), to address challenges in unsupervised person Re-ID.
To mitigate noise from camera variations and improve the accuracy and robustness of Re-ID models.

Main Methods:

Introduced Dynamic Adaptive Camera Jaccard (DACJ) distance to dynamically estimate and mitigate camera variations.
Proposed a Dynamic Adaptive Proxies (DAP) module with Dynamic Optimal Cluster Proxies (DOCP) and Dynamic Instance Proxies (DIP) for improved proxy construction.

Keywords:

Cluster proxies Contrastive learning Pseudo-label Unsupervised person re-identification

More Related Videos

Memorization-Based Training and Testing Paradigm for Robust Vocal Identity Recognition in Expressive Speech Using Event-Related Potentials Analysis

Memorization-Based Training and Testing Paradigm for Robust Vocal Identity Recognition in Expressive Speech Using Event-Related Potentials Analysis

Published on: August 9, 2024

Related Experiment Videos

Last Updated: Jun 20, 2026

Artificial Intelligence-Based System for Detecting Attention Levels in Students

Artificial Intelligence-Based System for Detecting Attention Levels in Students

Published on: December 15, 2023

Memorization-Based Training and Testing Paradigm for Robust Vocal Identity Recognition in Expressive Speech Using Event-Related Potentials Analysis

Memorization-Based Training and Testing Paradigm for Robust Vocal Identity Recognition in Expressive Speech Using Event-Related Potentials Analysis

Published on: August 9, 2024

Implemented a Dynamic Adaptive Knowledge Distillation (DAKD) module to generate refined soft labels.

Main Results:

The DACJ distance effectively reduces noise caused by camera variations during training.
The DAP module, leveraging DOCP and DIP, enhances clustering accuracy and proxy representation.
DAKD module generates improved soft labels, leading to more robust and accurate Re-ID performance.

Conclusions:

DAMCL framework demonstrates significant improvements in unsupervised person Re-ID.
The proposed dynamic adaptive modules effectively handle camera variations and noisy pseudo-labels.
Experimental results validate the efficiency and robustness of the DAMCL approach for person Re-ID tasks.