Nutrilens: visual calorie estimation with multi-class food detection using deep learning

With increasing awareness towards a healthy lifestyle, the use of digital tools to track daily diet and calorie intake is becoming more popular. To date, most food tracking applications are still based on manual input into the application, which can be tedious, often over-relying on a user’s memory, and messy. In order to overcome these challenges the NutriLens system has been proposed to automatically estimate calorie values from an image using deep learning. This proposed approach utilizes together multi-class food detection and estimation of portion size utilizing YOLOv11 and a reference object for accurately scaling real-world sizes. Each detected food item is matched to a %based food nutrition database to obtain a calorie value once detected from the image. The NutriLens system demonstrated a mean average precision (mAP@50) of 80.8%, which is higher than existing technology and is a strong indication of the accuracy and reliability of the system when estimating caloric content of diverse and complex Indian meals.Keywords: Calorie Estimation, Deep Learning, CNN, YOLOv11, Nutrition Analysis.