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 28, 2026

Artificial Intelligence-Based System for Detecting Attention Levels in Students
06:37

Artificial Intelligence-Based System for Detecting Attention Levels in Students

Published on: December 15, 2023

Harmonizing Scale for Intelligent Sensors: Resolution-Conditioned Adaptation of Vision Foundation Models for Crowd

Huan Xu1, Zhiheng Chen1, Sirou Shen1

  • 1School of Computer Engineering, Jimei University, Xiamen 361021, China.

Sensors (Basel, Switzerland)
|May 27, 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

Integrative single-cell and bulk transcriptomics define cell death patterns and ZDHHC22 in gastric cancer progression.

BMC medical genomics·2026
Same author

Enhancer-mediated Etv4 activation stimulates osteogenic differentiation.

Nature communications·2026
Same author

Targeted H<sub>2</sub>S Delivery System Attenuates Blood-Spinal Cord Barrier Disruption after Spinal Cord Injury by Reshaping the Ferritinophagy Pathway.

Advanced science (Weinheim, Baden-Wurttemberg, Germany)·2026
Same author

LLM-based medical dialogue dataset generation with automated instructions.

Scientific reports·2026
Same author

Day 3 embryo quality predicts clinical pregnancy and live birth rates following single frozen-thawed poor-quality blastocyst transfer.

International journal of gynaecology and obstetrics: the official organ of the International Federation of Gynaecology and Obstetrics·2026
Same author

Exogenous Melatonin Enhances the Salt Tolerance of Celery (<i>Apium graveolens</i> L.) by Regulating Osmotic Adaptation and Energy Metabolism via Starch and Sucrose Metabolic Pathways.

International journal of molecular sciences·2026

This study introduces D3-CalibCount, a new framework that adapts powerful DINOv3 foundation models for accurate crowd counting and localization, even with significant scale variations in surveillance footage.

Area of Science:

  • Computer Vision
  • Artificial Intelligence
  • Machine Learning

Background:

  • Scale variation poses a significant challenge for crowd-counting and localization in intelligent surveillance.
  • Conventional Convolutional Neural Network (CNN) backbones have limited representation capacity for complex crowd analysis.
  • Foundation models like DINOv3 offer strong self-supervised representations but require adaptation for density estimation and scale-aware tasks.

Purpose of the Study:

  • To propose D3-CalibCount, a parameter-efficient framework for adapting frozen DINOv3 representations for selective scale-aware crowd counting and localization.
  • To develop a lightweight Scale Harmonization Adapter (SHA) for calibrating DINOv3 features to be scale-selective.
  • To enable progressive feature inheritance across resolution levels for improved crowd density estimation.
Keywords:
crowd countingfoundation modelsscale variationselective inheritance learning

More Related Videos

Visualizing Visual Adaptation
04:43

Visualizing Visual Adaptation

Published on: April 24, 2017

Related Experiment Videos

Last Updated: May 28, 2026

Artificial Intelligence-Based System for Detecting Attention Levels in Students
06:37

Artificial Intelligence-Based System for Detecting Attention Levels in Students

Published on: December 15, 2023

Visualizing Visual Adaptation
04:43

Visualizing Visual Adaptation

Published on: April 24, 2017

Main Methods:

  • Utilized frozen DINOv3 representations as a backbone.
  • Introduced a trainable Scale Harmonization Adapter (SHA) for resolution-conditioned feature calibration.
  • Adapted generic DINOv3 features into scale-selective counting features for progressive inheritance.

Main Results:

  • D3-CalibCount demonstrated consistent improvements over selective inheritance baselines on three benchmarks.
  • The framework showed particular effectiveness in scenarios with severe scale variation.
  • Achieved reduced trainable parameters compared to traditional methods, though DINOv3 backbone increases inference cost.

Conclusions:

  • Lightweight adaptation of frozen foundation features is a practical approach for crowd counting and dense prediction tasks.
  • D3-CalibCount offers an efficient method for enhancing crowd analysis in intelligent surveillance.
  • The proposed framework balances performance gains with parameter efficiency for practical deployment.