Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Related Concept Videos

Vision

Vision

Vision is the result of light being detected and transduced into neural signals by the retina of the eye. This information is then further analyzed and interpreted by the brain. First, light enters the front of the eye and is focused by the cornea and lens onto the retina—a thin sheet of neural tissue lining the back of the eye. Because of refraction through the convex lens of the eye, images are projected onto the retina upside-down and reversed.

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

CoNav: Collaborative Cross-Modal Reasoning for Embodied Navigation.

IEEE transactions on pattern analysis and machine intelligence·2026

Same author

Role of NADPH oxidase 2-derived reactive oxygen species in cardiac electrophysiological disorders.

Channels (Austin, Tex.)·2026

Same author

Cyber Defense Effectiveness Evaluation for ICS Under Uncertainty: A Dynamic Bayesian Network Approach with Information Entropy.

Entropy (Basel, Switzerland)·2026

Same author

vNOTES Pelvic Reconstruction and Presacral-Uterosacral Ligament Compound Suspension for Treatment of Multicompartment Pelvic Organ Prolapse: A Three-Year, Three-Arm, Open-Label, Randomized Controlled Trial.

Journal of minimally invasive gynecology·2026

Same author

Bridging Subjectivity in Affective Explanation Captioning via Consensus-Prompted Emotion Reasoning.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026

Same author

Looking Broader for Knowledge Distillation Via Receptive-Field Alignment.

IEEE transactions on pattern analysis and machine intelligence·2026

Search research articles

Related Experiment Video

Updated: May 24, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

DHVT: Dynamic Hybrid Vision Transformer for Small Dataset Recognition.

Zhiying Lu, Chuanbin Liu, Xiaojun Chang

IEEE Transactions on Pattern Analysis and Machine Intelligence

|March 3, 2025

Summary

This study introduces the Dynamic Hybrid Vision Transformer (DHVT) and DHVT2, enhancing Vision Transformers (ViTs) for better image recognition. These models bridge the performance gap with Convolutional Neural Networks (CNNs), especially on limited datasets.

More Related Videos

Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing

Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing

Published on: February 23, 2024

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

Related Experiment Videos

Last Updated: May 24, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing

Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing

Published on: February 23, 2024

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

Area of Science:

Computer Science
Artificial Intelligence
Machine Learning

Background:

Vision Transformers (ViTs) exhibit a performance gap compared to Convolutional Neural Networks (CNNs) when trained on limited datasets due to insufficient inductive bias.
Key limitations in ViTs include inadequate spatial relevance and suboptimal channel representation, hindering their ability to capture fine-grained features and robust data patterns.

Purpose of the Study:

To address the limitations of ViTs by proposing the Dynamic Hybrid Vision Transformer (DHVT) and its computationally efficient variant, DHVT2.
To enhance spatial feature extraction and improve channel representation within the Vision Transformer architecture.

Main Methods:

DHVT integrates convolutional operations into the feature embedding and projection phases to bolster spatial relevance.
A dynamic aggregation mechanism and a novel 'head token' are employed to recalibrate and harmonize channel representations.
The study explores optimal network meta-structures, adopting a multi-stage hybrid design without a class token, further refined by a dimensional variable residual connection mechanism in DHVT2.

Main Results:

DHVT and DHVT2 achieve state-of-the-art results in image recognition tasks, effectively narrowing the performance disparity between ViTs and CNNs.
The proposed models demonstrate strong generalization capabilities in downstream experiments.

Conclusions:

The Dynamic Hybrid Vision Transformer (DHVT) and DHVT2 successfully overcome the inherent limitations of standard ViTs, particularly in data-scarce scenarios.
These hybrid architectures offer a promising direction for advancing vision-based artificial intelligence, providing competitive performance and enhanced generalization.