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: Apr 15, 2026

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

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

10.2K

Analysis of hyperparameter optimization effects on lightweight deep models for real-time image classification.

Vineet Kumar Rakesh1,2, Soumya Mazumdar3, Tapas Samanta4,5

  • 1Department of Engineering Sciences, Homi Bhabha National Institute, 400094, Mumbai, India. vineet@vecc.gov.in.

Scientific Reports
|April 13, 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

Neural Network-Based LoRa Received Signal Strength Indicator Fingerprint Identification for Indoor Localization of Mobile Robots.

Sensors (Basel, Switzerland)·2026
Same journal

Application of ephrin-B2 loaded glycol chitosan-silk fibroin hydrogel in the treatment of diabetic refractory wounds.

Scientific reports·2026
Same journal

International expert Delphi consensus on thromboprophylaxis in metabolic and bariatric surgery.

Scientific reports·2026
Same journal

Assessing the cross-region knowledge transfer capability of selected deep learning building vectorization methods in the context of available training datasets.

Scientific reports·2026
Same journal

Feasibility and preliminary effects of outdoor versus indoor cognitive-motor therapy in women with Alzheimer's disease: A randomized single-blind pilot study.

Scientific reports·2026
Same journal

Hallmarks of social action in the vocal turn-taking of wild common marmosets (Callithrix jacchus).

Scientific reports·2026
Same journal

Role and mechanism of AOPPs-induced NOX4-mediated ferroptosis in intervertebral disc degeneration.

Scientific reports·2026
See all related articles
This summary is machine-generated.

Optimizing hyperparameters for lightweight neural networks significantly improves accuracy and deployability on resource-constrained devices. Careful tuning enhances real-time image classification performance without changing model architecture.

Area of Science:

  • Computer Vision
  • Machine Learning
  • Deep Learning

Background:

  • Lightweight convolutional and transformer networks are crucial for real-time image classification on hardware with limited resources.
  • Their performance is highly sensitive to training hyperparameter choices, impacting practical applications.

Purpose of the Study:

  • To systematically analyze how hyperparameter tuning affects accuracy and deployability of modern lightweight neural networks.
  • To benchmark performance across various hardware platforms, including GPUs and edge CPUs.

Main Methods:

  • Seven lightweight backbones (ConvNeXt-Tiny, EfficientNetV2-S, MobileNetV3-L, MobileViT v2, RepVGG-A2, TinyViT-21M) were trained on ImageNet-1K.
  • Hyperparameters studied included learning rate, optimizers (SGD, AdamW), and regularization techniques (RandAugment, Mixup, CutMix, label smoothing).

Related Experiment Videos

Last Updated: Apr 15, 2026

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

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

10.2K
  • Automated hyperparameter searches (Optuna, population-based training) and deployment-focused evaluations (latency, throughput) were conducted.

    • Main Results:

    • Hyperparameter optimization without architectural changes led to consistent accuracy gains (0.5–3.5% Top-1).
    • MobileNetV3-L and RepVGG-A2 demonstrated excellent real-time performance (low latency, high throughput) on GPUs.
    • Edge CPU tests revealed minimal benefits from batching and underscored the importance of latency-focused model selection.

    Conclusions:

    • Controlled hyperparameter tuning is a cost-effective method to boost lightweight network performance for real-time image classification.
    • Model selection should consider specific hardware constraints, prioritizing latency for edge deployments.
    • The study provides insights into architecture-dependent stability and optimal operating points for various lightweight models.