Saved in:
Bibliographic Details
Main Authors: Amangeldi, Aidar, Taigonyrov, Angsar, Jawad, Muhammad Huzaifa, Mbonu, Chinedu Emmanuel
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.08259
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915795156598784
author Amangeldi, Aidar
Taigonyrov, Angsar
Jawad, Muhammad Huzaifa
Mbonu, Chinedu Emmanuel
author_facet Amangeldi, Aidar
Taigonyrov, Angsar
Jawad, Muhammad Huzaifa
Mbonu, Chinedu Emmanuel
contents This study evaluates the trade-offs between convolutional and transformer-based architectures on both medical and general-purpose image classification benchmarks. We use ResNet-18 as our baseline and introduce a fine-tuning strategy applied to four Vision Transformer variants (Tiny, Small, Base, Large) on DermatologyMNIST and TinyImageNet. Our goal is to reduce inference latency and model complexity with acceptable accuracy degradation. Through systematic hyperparameter variations, we demonstrate that appropriately fine-tuned Vision Transformers can match or exceed the baseline's performance, achieve faster inference, and operate with fewer parameters, highlighting their viability for deployment in resource-constrained environments.
format Preprint
id arxiv_https___arxiv_org_abs_2505_08259
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CNN and ViT Efficiency Study on Tiny ImageNet and DermaMNIST Datasets
Amangeldi, Aidar
Taigonyrov, Angsar
Jawad, Muhammad Huzaifa
Mbonu, Chinedu Emmanuel
Computer Vision and Pattern Recognition
This study evaluates the trade-offs between convolutional and transformer-based architectures on both medical and general-purpose image classification benchmarks. We use ResNet-18 as our baseline and introduce a fine-tuning strategy applied to four Vision Transformer variants (Tiny, Small, Base, Large) on DermatologyMNIST and TinyImageNet. Our goal is to reduce inference latency and model complexity with acceptable accuracy degradation. Through systematic hyperparameter variations, we demonstrate that appropriately fine-tuned Vision Transformers can match or exceed the baseline's performance, achieve faster inference, and operate with fewer parameters, highlighting their viability for deployment in resource-constrained environments.
title CNN and ViT Efficiency Study on Tiny ImageNet and DermaMNIST Datasets
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2505.08259