Food Calorie and Volume Estimation from Images Using YOLOv5

Abstract

Aktaruzzaman Siddiquei, Ahsanul Islam, Al-Amin Hossain, Sohag Hasan, Susmita Das, Sabbir Shikdar, Mehedi Hasan, Lubbabah Sugra Siddiqi Tamanna, K M Fysal Kabir, Nazmul Hossain, Nur-E-Iman Nasim Talukdar, Shariful Islam, Eurid Al Muttakim, Apple Sarker, Farjana Rahman, Madhobi Pramanik, Jannatul Ferdous Swarna and Israth Jahan Sonda

Accurate assessment of food intake is crucial for weight management and health. This report presents a deep learning approach leveraging YOLOv5, a state-of-the-art object detection model, to estimate food calories and amounts from images. The proposed workflow detects food items via a customized YOLOv5 model and refines segmentation masks using semantic segmentation. 3D shape recognition and reconstruction techniques estimate food volume, while integrated pre- trained ingredient classifiers and nutritional databases provide calorie information. Preliminary results on a benchmark food image dataset demonstrate the approach's ability to accurately quantify calories and portion sizes for complex meals. The system has the potential to assist consumers in making informed dietary choices and provide insights for public health initiatives related to nutrition.

PDF