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: Jul 13, 2026

End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
03:31

End-To-End Deep Neural Network for Salient Object Detection in Complex Environments

Published on: December 15, 2023

SynerNet: Broad-to-precise CAM synergy for weakly supervised semantic segmentation.

Zhonggai Wang1, Guangyu Gao1, Zhuoshu Li1

  • 1School of Computer Science and Technology, Beijing Institute of Technology, Beijing, China.

Neural Networks : the Official Journal of the International Neural Network Society
|May 6, 2026
PubMed
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

FGF21 rejuvenates aged human adipose-derived mesenchymal stem cells via enhancement of TFE3-mediated autophagy flux.

Autophagy·2026
Same author

Growth arrest-specific 6 rejuvenates senescent HUCMSCs through upregulating Nrf2 for diabetic wound therapy.

Free radical biology & medicine·2026
Same author

Associations of changes in frailty with multiple chronic diseases in middle aged and elderly adults.

BMC public health·2025
Same author

AutoSGRL: Automated framework construction for self-supervised graph representation learning.

Neural networks : the official journal of the International Neural Network Society·2025
Same author

DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning.

Nature·2025
Same author

Genetic Evidence on the Role of Blood Phytosterols in Frailty: A Mendelian Randomization Study.

Food science & nutrition·2025

SynerNet, a novel one-stage dual-branch framework, enhances weakly supervised semantic segmentation by generating both broad and precise pseudo-labels. This approach overcomes limitations of traditional methods, achieving state-of-the-art results.

Area of Science:

  • Computer Vision
  • Deep Learning
  • Image Segmentation

Background:

  • Weakly Supervised Semantic Segmentation (WSSS) is challenging due to noisy image-level labels.
  • One-stage WSSS methods suffer from encoder-label coupling, reinforcing initial inaccuracies.
  • Existing approaches struggle with incomplete or imprecise class activation maps (CAMs).

Purpose of the Study:

  • To introduce SynerNet, a one-stage dual-branch framework for improved WSSS.
  • To address the limitations of error propagation in coupled encoder-decoder architectures.
  • To generate high-fidelity pseudo-labels that are both comprehensive and accurate.

Main Methods:

  • Employed a dual-branch architecture with complementary objectives: B-CAM for broad coverage and P-CAM for precise localization.
Keywords:
CAMCo-trainingDual-branchPseudo-labelWeakly supervised semantic segmentation

More Related Videos

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

Related Experiment Videos

Last Updated: Jul 13, 2026

End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
03:31

End-To-End Deep Neural Network for Salient Object Detection in Complex Environments

Published on: December 15, 2023

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

  • Utilized global attention in B-CAM to expand foreground regions and emphasized background in P-CAM for localization.
  • Implemented cross-supervision between branches to decouple optimization and mitigate error reinforcement.
  • Introduced a confidence matrix from multi-scale ViT features for confidence-guided fusion of predictions.

    • Main Results:

    • SynerNet achieved state-of-the-art performance on PASCAL VOC 2012 and COCO 2014 datasets.
    • The dual-branch approach successfully generated pseudo-labels that were both complete and precise.
    • Demonstrated effective mitigation of error reinforcement through decoupled optimization and confidence-guided fusion.

    Conclusions:

    • SynerNet offers an effective one-stage co-training strategy for high-quality WSSS.
    • The proposed complementary objectives and confidence-guided fusion significantly improve segmentation accuracy.
    • The framework successfully generates high-fidelity pseudo-labels in an end-to-end manner.