An Experimental Study on Estimating Residual Quantity of Foodstuff based on Deep Learning using 3D Model

Abstract

With the recent development of deep learning techniques, many image recognition methods have been proposed for various objects including foods and ingredients. However, preparing training datasets of real objects is difficult, because foods and ingredients often deteriorate quickly and a number of those types are large. Therefore, we have been studying a deep learning-based image recognition method which uses 3D models as training dataset. In this paper, we focus on an estimation of remaining amount of food ingredients, and we conduct an experiment on a simple sphere as an initial step of our study, and report its result.