Comparative Analysis of VGG16 and ResNet-50 Architectures for Food Image Classification: A Deep Dive Into Performance and Practical Implications

This paper presents the findings of a computer vision project aimed at comparing the performance of two Convolutional Neural Network (CNN) architectures, VGG16 and ResNet-50, on the Food-101 dataset. The primary objectives were to compare the efficacy of these models in food image classification and determine which architecture yields superior performance. The methodology involved training both models on the Food-101 dataset and assessing their accuracy metrics. Findings indicate that ResNet-50 outperformed VGG16 in terms of classification accuracy. Consequently, ResNet-50 was selected for further experimentation, which revealed the model's competency in classifying a collection of unseen food images, accurately identifying most of the images from 101 food classes. The conclusions drawn from this paper affirm the superiority of ResNet-50 over VGG16 for food image classification tasks, achieving an average accuracy of 71% compared to VGG16's 33% on the test set. This disparity underscores ResNet-50's effectiveness in real-world applications, demonstrating its robust performance in food image classification.