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 12, 2026

Quantifying Intermembrane Distances with Serial Image Dilations
07:45

Quantifying Intermembrane Distances with Serial Image Dilations

Published on: September 28, 2018

ELBO-T2IAlign: A Generic ELBO-Based Method for Calibrating Pixel-level Text-Image Alignment in Diffusion Models.

Qin Zhou, Zhiyang Zhang, Jinglong Wang

    IEEE Transactions on Image Processing : a Publication of the IEEE Signal Processing Society
    |July 9, 2026
    PubMed
    Summary
    This summary is machine-generated.

    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

    MUFOLD-DB: a processed protein structure database for protein structure prediction and analysis.

    BMC genomics·2015
    Same author

    The I-TASSER Suite: protein structure and function prediction.

    Nature methods·2014
    Same author

    Genome-wide expression analysis of soybean NF-Y genes reveals potential function in development and drought response.

    Molecular genetics and genomics : MGG·2014
    Same author

    Classification of lung cancer using ensemble-based feature selection and machine learning methods.

    Molecular bioSystems·2014
    Same author

    Resveratrol possesses protective effects in a pristane-induced lupus mouse model.

    PloS one·2014
    Same author

    Protein-losing enteropathy in systemic lupus erythematosus: 12 years experience from a Chinese academic center.

    PloS one·2014

    Diffusion models can generate images but struggle with perfect text-image alignment, especially for small or rare objects. A new training-free method, ELBO-T2IAlign, calibrates this alignment for improved downstream tasks.

    Area of Science:

    • Artificial Intelligence
    • Computer Vision
    • Machine Learning

    Background:

    • Diffusion models generate high-quality images and encode text-image alignment.
    • Current methods assume perfect alignment, which is often not true.
    • Pixel-level and class-level text-image misalignment is a significant issue.

    Purpose of the Study:

    • Evaluate pixel-level image and class-level text alignment in diffusion models.
    • Analyze the causes of misalignment, particularly training data bias.
    • Propose a method to calibrate text-image alignment in diffusion models.

    Main Methods:

    • Utilized zero-shot referring image segmentation as a proxy task.
    • Analyzed misalignment in diffusion models concerning training data bias.

    More Related Videos

    Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration
    07:03

    Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration

    Published on: February 23, 2017

    Related Experiment Videos

    Last Updated: Jul 12, 2026

    Quantifying Intermembrane Distances with Serial Image Dilations
    07:45

    Quantifying Intermembrane Distances with Serial Image Dilations

    Published on: September 28, 2018

    Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration
    07:03

    Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration

    Published on: February 23, 2017

  • Introduced ELBO-T2IAlign, a training-free calibration method based on evidence lower bound (ELBO).

    • Main Results:

    • Identified misalignment issues in images with small-sized, occluded, or rare object classes.
    • Demonstrated that ELBO-T2IAlign effectively calibrates pixel-text alignment.
    • Showcased improvements in downstream tasks like segmentation and image editing.

    Conclusions:

    • ELBO-T2IAlign is a generic and effective method for improving text-image alignment in diffusion models.
    • The training-free approach requires no additional annotations or retraining.
    • Calibration enhances performance across various complementary downstream applications.