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: Jun 27, 2026

Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

Model architecture dominates nutritional estimation accuracy in vision-language systems.

Luca Vedovelli1,2, Sofia Pugnaloni1,3, Corrado Lanera1,2

  • 1Unit of Biostatistics, Epidemiology, and Public Health, Department of Cardiac, Thoracic, Vascular Sciences and Public Health, University of Padua, Padua, Italy.

Scientific Reports
|June 25, 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

Recognition of eating episodes via commercial smartwatch sensors analysis.

PLOS digital health·2026
Same author

Enhancing Hospital Nutrition Assessment Through Artificial Intelligence: A Prospective Tray-Level Pilot Study.

Nutrients·2026
Same author

Wernicke Encephalopathy Complicating a Distinctive POLG Phenotype With MNGIE-Like Features.

European journal of neurology·2026
Same author

[Occupational cancers: open issues and perspectives in cancer research, surveillance, detection, and prevention tools].

Epidemiologia e prevenzione·2026
Same author

A pipeline for developing deep learning prognostic prediction models in cardiac magnetic resonance image analysis.

European heart journal. Digital health·2026
Same author

Genome-wide association for sarcoidosis identifies novel risk loci and genetic heritability in African and European ancestries: a meta-analysis from the Finngen, Million Veteran Program, UK Biobank, and Biobank Japan datasets.

Orphanet journal of rare diseases·2025

Vision-language models (VLMs) show promise for automated dietary assessment. Model architecture and image quality are key, not multiple angles or complex prompts. Current VLMs are best for calorie tracking, not clinical nutrition analysis.

Area of Science:

  • Artificial Intelligence
  • Computer Vision
  • Nutritional Science

Background:

  • Vision-language models (VLMs) offer potential for automated dietary assessment using food images.
  • Optimal strategies for deploying VLMs in dietary assessment are not yet established.

Purpose of the Study:

  • To evaluate the impact of different VLM architectures, image configurations, prompt strategies, and image quality on automated dietary assessment accuracy.
  • To benchmark VLM performance against professional nutritionists.

Main Methods:

  • Evaluated 40 VLMs from eight providers using the Nutrition5k database and original food images.
  • Tested image configurations (1-5 angles), prompt strategies (7 approaches), and image quality (consumer smartphone vs. laboratory imaging).
  • Compared VLM performance to professional nutritionists' assessments.
Keywords:
Automated nutrition analysisComputer visionDeep learningDietary assessmentFood recognitionNutritional estimationPrompt engineeringVision-language models

Related Experiment Videos

Last Updated: Jun 27, 2026

Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

Main Results:

  • VLM architecture was the primary driver of performance variance (99.6%).
  • High-quality consumer smartphone images outperformed controlled laboratory images (RMSLE 0.548 vs. 0.616).
  • Professional nutritionists significantly outperformed all VLMs (RMSLE 0.176 vs. 0.443), especially for protein estimation.

Conclusions:

  • Model selection and image quality are critical for nutritional estimation accuracy in VLMs.
  • Multi-angle imaging and prompt complexity have negligible effects on VLM performance.
  • Current VLMs are suitable for consumer calorie tracking but require architectural advancements for clinical-grade macronutrient profiling.