Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Dec 15, 2025

Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

9.8K

Weakly Supervised Person Re-ID: Differentiable Graphical Learning and a New Benchmark.

Guangrun Wang, Guangcong Wang, Xujie Zhang

    IEEE Transactions on Neural Networks and Learning Systems
    |July 15, 2020
    PubMed
    Summary

    This study introduces weakly supervised person reidentification (Re-ID) using bag-level annotations, significantly reducing labeling costs. A novel differentiable graphical model generates reliable pseudolabels for training Re-ID models effectively.

    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

    Agentic Designer: Progressive Multi-Agent Collaboration for Structure-Aware Interior Layout Generation.

    IEEE transactions on pattern analysis and machine intelligence·2026
    Same author

    Embodied cognition-driven interpretable trajectory prediction of autonomous systems.

    Nature communications·2026
    Same author

    Occupational health risks and abatement trade-offs of VOCs in key industries of the Yangtze River Delta: Insights for a shift from concentration-based compliance to reactivity- and toxicity-driven management.

    Journal of hazardous materials·2026
    Same author

    Fast formation to reinforce lithium-rich cathodes.

    Nature·2026
    Same author

    Comparison of the predictive value of CONUT, NLR, and PNI for 6-month and 1-year mortality in middle-aged and older adults with hip fractures: a retrospective study.

    Frontiers in nutrition·2026
    Same author

    Benefiting From OOD Samples in Open-Set Semi-Supervised Object Detection.

    IEEE transactions on neural networks and learning systems·2026

    Area of Science:

    • Computer Vision
    • Machine Learning
    • Artificial Intelligence

    Background:

    • Accurate annotations for person reidentification (Re-ID) datasets like CUHK03 and Market-1501 are costly and time-consuming.
    • Existing Re-ID benchmarks are limited in scale, hindering the development of robust models.

    Purpose of the Study:

    • To reduce the annotation effort for Re-ID datasets by introducing bag-level labels instead of image-level labels.
    • To create a large-scale Re-ID benchmark, SYSU-30k, to facilitate research in weakly supervised Re-ID.
    • To develop a novel method for learning Re-ID models from weakly supervised data.

    Main Methods:

    • Introduced a weakly supervised learning paradigm for Re-ID using bag-level annotations.
    • Created the SYSU-30k benchmark, containing 30,000 individuals and over 29 million images.

    Related Experiment Videos

    Last Updated: Dec 15, 2025

    Deep Neural Networks for Image-Based Dietary Assessment
    13:19

    Deep Neural Networks for Image-Based Dietary Assessment

    Published on: March 13, 2021

    9.8K
  • Developed a differentiable graphical model to capture inter-image dependencies within bags and generate pseudolabels for training.

    • Main Results:

    • Achieved state-of-the-art performance on the SYSU-30k benchmark using the proposed weakly supervised method.
    • Demonstrated the effectiveness of the differentiable graphical model in generating reliable pseudolabels.
    • The proposed method shows competitive results compared to fully supervised Re-ID models on various datasets.

    Conclusions:

    • Weakly supervised Re-ID with bag-level annotations is a viable and cost-effective alternative to fully supervised methods.
    • The SYSU-30k benchmark provides a valuable resource for advancing research in large-scale, weakly supervised Re-ID.
    • The differentiable graphical model offers a promising approach for learning from noisy or incomplete annotations in Re-ID.