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

MPQ-DMv2: Flexible Residual Mixed Precision Quantization for Low-Bit Diffusion Models With Temporal Distillation.

IEEE transactions on pattern analysis and machine intelligence·2026
Same author

The Mediating Role of Self-efficacy in Acute Myocardial Infarction Patients between Post-PCI Fatigue and Post-traumatic Stress Disorder.

Alternative therapies in health and medicine·2024
Same author

Sequential Point Clouds: A Survey.

IEEE transactions on pattern analysis and machine intelligence·2024
Same author

Robust and accurate pulmonary nodule detection with self-supervised feature learning on domain adaptation.

Frontiers in radiology·2023
Same author

AI-Driven Robust Kidney and Renal Mass Segmentation and Classification on 3D CT Images.

Bioengineering (Basel, Switzerland)·2023
Same author

Virtual contrast enhancement for CT scans of abdomen and pelvis.

Computerized medical imaging and graphics : the official journal of the Computerized Medical Imaging Society·2022

Related Experiment Video

Updated: Dec 11, 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

876

Unambiguous Text Localization, Retrieval, and Recognition for Cluttered Scenes.

Xuejian Rong, Chucai Yi, Yingli Tian

    IEEE Transactions on Pattern Analysis and Machine Intelligence
    |August 22, 2020
    PubMed
    Summary

    This study introduces new models for accurately locating, retrieving, and recognizing specific text in cluttered images. These methods improve scene understanding by precisely identifying text instances from descriptions.

    More Related Videos

    Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
    08:25

    Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment

    Published on: May 7, 2019

    9.4K
    Photorealistic Learned Landscapes for Augmented Reality
    06:54

    Photorealistic Learned Landscapes for Augmented Reality

    Published on: June 27, 2025

    561

    Related Experiment Videos

    Last Updated: Dec 11, 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

    876
    Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
    08:25

    Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment

    Published on: May 7, 2019

    9.4K
    Photorealistic Learned Landscapes for Augmented Reality
    06:54

    Photorealistic Learned Landscapes for Augmented Reality

    Published on: June 27, 2025

    561

    Area of Science:

    • Computer Vision
    • Artificial Intelligence
    • Natural Language Processing

    Background:

    • Scene text understanding is crucial for AI, but current methods struggle with cluttered environments and specific text instances.
    • Existing visual phrase grounding often focuses on general objects, neglecting precise text localization and recognition.

    Purpose of the Study:

    • To develop advanced models for accurate scene text localization, retrieval, and recognition from natural language descriptions.
    • To address the challenge of identifying specific text instances within complex and crowded visual scenes.

    Main Methods:

    • A novel recurrent dense text localization network (DTLN) for decoding image features into distinct text detections, avoiding redundant detections.
    • A context reasoning text retrieval (CRTR) model that jointly encodes text instances and context for improved bounding box ranking.
    • A recurrent text recognition module to enhance the models' capabilities through text verification or transcription.

    Main Results:

    • The proposed DTLN effectively handles crowded text instances and avoids scale-based repetitions.
    • The CRTR model successfully ranks localized text bounding boxes using context compatibility.
    • Evaluations on benchmark datasets demonstrate the models' effectiveness in joint scene text localization, retrieval, and recognition.

    Conclusions:

    • The developed models offer significant improvements for the integrated task of scene text localization, retrieval, and recognition.
    • These advancements contribute to a more robust understanding of complex visual scenes through precise text analysis.