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 Videos

Benchmarking a Multimodal and Multiview and Interactive Dataset for Human Action Recognition.

An-An Liu, Ning Xu, Wei-Zhi Nie

    IEEE Transactions on Cybernetics
    |July 19, 2016
    PubMed
    Summary

    A new multimodal, multiview, and interactive (M2I) dataset enables comprehensive evaluation of human action recognition across single-view, cross-view, cross-domain, and multitask learning scenarios. This dataset addresses limitations in existing benchmarks for robust algorithm comparison.

    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

    Programmable In Vivo Synthesis of Quantum Dots.

    Angewandte Chemie (International ed. in English)·2026
    Same author

    SpineVLM: A Markdown-Guided Structured Fine-Tuning Framework for Spine X-ray Report Generation.

    IEEE journal of biomedical and health informatics·2026
    Same author

    Simultaneous Localization of Electrons/Holes by Surface Cationic Dangling-Bonds/Vacancies for Synergistically Boosting Photocatalytic Activity.

    Journal of the American Chemical Society·2026
    Same author

    Biologically Adaptable Quantum Dots: Intracellular in Situ Synthetic Strategy and Mechanism.

    Accounts of chemical research·2026
    Same author

    Free-VTON: Cost-free acceleration and quality enhancement for diffusion-based virtual try-on.

    Neural networks : the official journal of the International Neural Network Society·2026
    Same author

    Proton-Resistant Quantum Dots by Ligands.

    ACS nano·2025

    Area of Science:

    • Computer Vision
    • Machine Learning
    • Human Action Recognition

    Background:

    • Human action recognition is a key research area with evolving learning paradigms: single-view, cross-view, cross-domain, and multitask learning.
    • Existing datasets often support only a subset of these learning problems, limiting algorithm comparison due to dataset and experimental variances.
    • There is a lack of a unified dataset for concurrent analysis across all four major human action recognition learning problems.

    Purpose of the Study:

    • Introduce the novel multimodal, multiview, and interactive (M2I) dataset.
    • Design M2I for comprehensive evaluation of human action recognition methods under single-view, cross-view, cross-domain, and multitask learning.
    • Provide a benchmark for comparing state-of-the-art algorithms across diverse learning scenarios.

    Related Experiment Videos

    Main Methods:

    • Developed the M2I dataset comprising 1760 action samples across 22 categories, including person-person and person-object interactions.
    • Systematically benchmarked 13 state-of-the-art approaches using nine feature and descriptor combinations on the M2I dataset.
    • Evaluated algorithm performance across single-view, cross-view, cross-domain, and multitask learning problems.

    Main Results:

    • The M2I dataset presents significant challenges, including high intraclass variation and view differences.
    • Multiple action categories exhibit high similarity, further complicating recognition tasks.
    • Benchmarking revealed varying performance of state-of-the-art methods across the four learning problems.

    Conclusions:

    • The M2I dataset offers a robust foundation for evaluating human action recognition algorithms.
    • Its design facilitates concurrent analysis of diverse learning problems, addressing limitations of prior datasets.
    • The comprehensive benchmarking provides insights into the strengths and weaknesses of current approaches.