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: May 20, 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

442

FDC: Feature Dropout Consistency for unsupervised domain adaptation semantic segmentation.

Chaoyu Rao1, Wanshu Fan1, Cong Wang2

  • 1National and Local Joint Engineering Laboratory of Computer Aided Design, School of Software Engineering, Dalian University, China.

Neural Networks : the Official Journal of the International Neural Network Society
|May 17, 2025
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

Stationary state distribution and efficiency analysis of the Langevin equation via real or virtual dynamics.

The Journal of chemical physics·2017
Same author

Optimized production and isolation of antibacterial agent from marine <i>Aspergillus flavipes</i> against <i>Vibrio harveyi</i>.

3 Biotech·2017
Same author

Preparation and antioxidant properties of low molecular holothurian glycosaminoglycans by H<sub>2</sub>O<sub>2</sub>/ascorbic acid degradation.

International journal of biological macromolecules·2017
Same author

Associations between single-nucleotide polymorphisms in the NTRK1 gene and basal pain sensitivity in young Han Chinese women.

Neuroscience letters·2017
Same author

Design, synthesis and antiproliferative effect of 17β-amide derivatives of 2-methoxyestradiol and their studies on pharmacokinetics.

Steroids·2017
Same author

MiR-324-3p promotes tumor growth through targeting DACT1 and activation of Wnt/β-catenin pathway in hepatocellular carcinoma.

Oncotarget·2017
This summary is machine-generated.

Feature Dropout Consistency (FDC) prevents overfitting in unsupervised domain adaptation semantic segmentation by perturbing features. This novel approach enhances self-training methods, setting new benchmarks in domain adaptation tasks.

Area of Science:

  • Computer Vision
  • Machine Learning
  • Artificial Intelligence

Background:

  • Unsupervised Domain Adaptation Semantic Segmentation (UDASS) faces overfitting challenges due to the lack of target labels.
  • Self-training is effective but requires regularization to prevent model overfitting on unlabeled target data.
  • Consistency techniques regularize models by enforcing consistent predictions under data perturbations.

Purpose of the Study:

  • To introduce a novel Feature Dropout Consistency (FDC) module for UDASS.
  • To address the overfitting issue in self-training methods for semantic segmentation.
  • To enhance model robustness by introducing perturbations at the feature level.

Main Methods:

  • Implemented random feature dropout between the encoder and decoder of the student network.
Keywords:
Consistency techniquePerturbationSelf-trainingSemantic segmentationUnsupervised domain adaptation

More Related Videos

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
05:56

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application

Published on: April 14, 2023

2.3K
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

335

Related Experiment Videos

Last Updated: May 20, 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

442
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
05:56

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application

Published on: April 14, 2023

2.3K
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

335
  • Introduced a feature dropout consistency loss to minimize prediction discrepancies between perturbed and unperturbed networks.
  • Integrated FDC with existing self-training methodologies and explored combined input and feature perturbations.

    • Main Results:

    • FDC consistently outperformed baseline models in standard UDA settings.
    • The method achieved new benchmark performance on GTAV → Cityscapes and SYNTHIA → Cityscapes datasets.
    • Feature-level perturbations proved effective in regularizing models and improving adaptation.

    Conclusions:

    • Feature Dropout Consistency is a highly effective module for unsupervised domain adaptation semantic segmentation.
    • FDC enhances self-training by introducing feature-level regularization, mitigating overfitting.
    • The proposed method sets new performance standards for cross-domain semantic segmentation tasks.