Answers are grounded in this paper and cite their sources.
Automated food image classification is a significant research problem given the large number of food-related photographs flooding the social media and the internet. Food image classification is mainly practiced with the help of Convolutional Neural Networks (CNNs) having many hidden layers that are known as deep CNNs. The deep pre-trained CNN architectures have evolved over the past few years; they operate on the concept of transfer learning of knowledge from one domain to the other. In this paper, we have analyzed the performances of five popular very deep pre-trained networks namely, Inception-v3 with 48 layers, EfficientNet-B0 with 237 layers, Xception with 71 layers, DenseNet-121 with 121 layers, and MobileNet with 53 layers, for the classification of food images from the benchmark Food-101 dataset comprising of 101 food categories. Our experiments prove that Xception gives the best performance for classifying the 101 categories of food images, with an accuracy of 84.54%, significantly outperforming the other deep pre-trained networks.
No takes yet. Share an insight, caveat, or question.
Singh et al. (2023) studied this question.