Comparing Convolutional Neural Networks and Transformers in a Points-of-Interest Experiment

This paper addresses a research gap by providing a unique comparative analysis of the most prevalent Deep Learning (DL) models for image classification, specifically focusing on Points-of-Interest (POI) and discusses their differences. Convolutional Neural Network (CNN) based models are trained on a POI dataset and their accuracy levels are noted. The paper then proceeds to compare them with a recent model called ViT, which is based on the Transformers architecture, and has the potential to surpass current accuracy levels and bring further innovation in the field of Deep Learning. For this comparative study, a random sample from the Places365 dataset is utilized and is referred as the mini-places dataset in this paper.