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 Concept Videos

You might also read

Related Articles

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

Sort by
Same author

RAD51 gene is associated with advanced age-related macular degeneration in Chinese population.

Clinical biochemistry·2013
Same author

Immunization against recombinant GnRH-I alters ultrastructure of gonadotropin cell in an experimental boar model.

Reproductive biology and endocrinology : RB&E·2013
Same author

Multi-class constrained normalized cut with hard, soft, unary and pairwise priors and its applications to object segmentation.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2013
Same author

Comparison of genomic and amino acid sequences of eight Japanese encephalitis virus isolates from bats.

Archives of virology·2013
Same author

Regulation of dendritic cell differentiation in bone marrow during emergency myelopoiesis.

Journal of immunology (Baltimore, Md. : 1950)·2013
Same author

Separation of mandelic acid and its derivatives with new immobilized cellulose chiral stationary phase.

Journal of Zhejiang University. Science. B·2013

Related Experiment Video

Updated: Jul 12, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

565

Back to Reality: Learning Data-Efficient 3D Object Detector With Shape Guidance.

Xiuwei Xu, Ziwei Wang, Jie Zhou

    IEEE Transactions on Pattern Analysis and Machine Intelligence
    |October 31, 2023
    PubMed
    Summary

    This study introduces a weakly-supervised 3D object detection method using synthetic data to enhance limited position-level annotations. The approach significantly improves detection accuracy with minimal labeling effort.

    More Related Videos

    Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing
    06:25

    Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing

    Published on: February 23, 2024

    627
    Three-Dimensional Shape Modeling and Analysis of Brain Structures
    05:33

    Three-Dimensional Shape Modeling and Analysis of Brain Structures

    Published on: November 14, 2019

    7.1K

    Related Experiment Videos

    Last Updated: Jul 12, 2025

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
    03:31

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

    Published on: December 15, 2023

    565
    Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing
    06:25

    Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing

    Published on: February 23, 2024

    627
    Three-Dimensional Shape Modeling and Analysis of Brain Structures
    05:33

    Three-Dimensional Shape Modeling and Analysis of Brain Structures

    Published on: November 14, 2019

    7.1K

    Area of Science:

    • Computer Vision
    • Machine Learning
    • 3D Reconstruction

    Background:

    • 3D object detection typically requires extensive box-level annotations, which are labor-intensive to create.
    • Weakly-supervised methods offer a potential solution by utilizing less detailed annotations, but often suffer from performance limitations.
    • Existing methods struggle to bridge the information gap between sparse position-level labels and the dense information needed for robust 3D detection.

    Purpose of the Study:

    • To develop a novel weakly-supervised approach for 3D object detection that trains a strong detector using only position-level annotations (object centers and categories).
    • To overcome the inherent information loss from position-level annotations by leveraging synthetic data and domain adaptation techniques.
    • To significantly reduce the annotation effort required for training high-performance 3D object detection models.

    Main Methods:

    • A shape-guided label enhancement method is proposed to generate virtual scenes with box-level annotations from real-world position-level data.
    • Virtual-to-real domain adaptation is employed to transfer knowledge from synthetic scenes to refine real-world annotations and supervise detector training.
    • Differentiable label enhancement and a label-assisted self-training strategy are introduced to optimize virtual scenes and generate pseudo-box labels for fully-supervised training.

    Main Results:

    • The proposed weakly-supervised method achieves state-of-the-art performance on the ScanNet and Matterport3D datasets, outperforming existing weakly- and semi-supervised approaches.
    • The method demonstrates comparable detection performance to popular fully-supervised methods.
    • The approach requires less than 5% of the labeling labor compared to fully-supervised methods, showcasing significant efficiency gains.

    Conclusions:

    • The developed weakly-supervised 3D object detection framework effectively utilizes synthetic data and domain adaptation to compensate for limited annotations.
    • The method offers a practical and efficient solution for training accurate 3D detectors, significantly reducing annotation costs.
    • This work paves the way for more accessible and scalable 3D object detection in real-world applications.