Convolutional neural network and vision transformer for image classification
Jiaqi Lu · Applied and Computational Engineering · 2023
Visual Transformer (ViT) has been a hot topic for research for the past few years after it first emerged in the field. On image recognitions, due to the amount of information ViT could retrieve from the source image, in cases it can rival the traditionally prevailing Convolutional Neural Network (CNN). Then there emerged different models based on ViT, all being built having a specific field or a flaw not addressed by original ViT in mind. In this paper these models are being tested on the same dataset along with a standard CNN to see how they perform compare to each other, and the best performing ViT model was then changed to see how there would be some possible improvements.