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

Coordinated Subpocket Engagement Underlies Nitazene Potency at the μ-Opioid Receptor.

ACS chemical neuroscience·2026
Same author

Evaluating and Mitigating Carbon Dioxide Equivalent Emissions in Stroke Management: A Modeling Study.

Journal of the American Heart Association·2026
Same author

Polypill for heart failure with reduced ejection fraction: the POLY-HF randomized trial.

Nature medicine·2026
Same author

Resetting without resetting: An alternate strategy to experimentally verify optimal mean first passage time under stochastic resetting.

Physical review. E·2026
Same author

Heart Failure Duration, Cardiac Remodeling, Dysfunction, and Hemodynamic Severity in HFpEF and HFmrEF: Insights From REDUCE LAP-HF II.

JACC. Heart failure·2026
Same author

Inhibition of human FcγRIII prevents platelet destruction in a dual-humanized murine model of FNAIT.

Blood advances·2026

Related Experiment Video

Updated: Jun 21, 2025

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

385

An UltraMNIST classification benchmark to train CNNs for very large images.

Deepak K Gupta1,2, Udbhav Bamba1, Abhishek Thakur3

  • 1Transmute AI Lab (Texmin Hub), Indian Institute of Technology, ISM Dhanbad, India.

Scientific Data
|July 12, 2024
PubMed
Summary

A new UltraMNIST dataset and benchmarks are introduced for training convolutional neural networks (CNNs) on large scientific images. This resource aims to advance CNN development for complex, multi-scale data challenges in fields like satellite imaging.

More Related Videos

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

2.7K
Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

9.0K

Related Experiment Videos

Last Updated: Jun 21, 2025

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

385
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

2.7K
Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

9.0K

Area of Science:

  • Computer Vision
  • Machine Learning
  • Scientific Data Analysis

Background:

  • Existing convolutional neural networks (CNNs) are ill-suited for large-scale scientific images (e.g., satellite, microscopy) due to their complex multi-scale features.
  • A critical need exists for specialized CNN architectures and application-independent datasets to address the unique challenges of processing vast scientific imagery.
  • Current research lacks high-quality, challenging datasets essential for developing and validating novel CNN approaches for large image analysis.

Purpose of the Study:

  • To introduce the 'UltraMNIST dataset' and associated benchmarks, specifically designed for the research problem of training CNNs for large scientific images.
  • To provide a flexible and representative dataset that facilitates the development of new CNN methods capable of handling multi-scale features and varying complexities.
  • To spur innovation in CNN architectures optimized for large image processing, addressing both high-performance and memory-constrained scenarios.

Main Methods:

  • Development of the 'UltraMNIST dataset', a customizable benchmark representative of scientific data challenges.
  • Introduction of two problem variants: a standard version for optimal GPU resource utilization and a budget-aware version for constrained memory environments.
  • Presentation of baseline CNN models and analysis of the impact of reduced image resolution on performance.

Main Results:

  • The UltraMNIST dataset offers a scalable and adaptable platform for evaluating CNNs on large scientific images.
  • Baseline performance metrics are established, providing a foundation for future research and method development.
  • The study investigates the trade-offs between image resolution, computational resources, and CNN performance.

Conclusions:

  • The UltraMNIST dataset and benchmarks are expected to accelerate the development of advanced CNNs tailored for large-scale scientific image analysis.
  • This resource will enable researchers to explore novel CNN architectures and training strategies for domains like remote sensing and digital pathology.
  • The findings highlight the importance of specialized datasets and benchmarks in driving progress in AI for scientific discovery.