FoodVision

Project Overview

FoodVision Demo

About FoodVision

FoodVision is an advanced deep learning-based food detection system that leverages the power of YOLOv8 to identify and classify various food items in images. The system is designed to detect 55 different food classes, with a particular focus on fruits and vegetables, making it a valuable tool for dietary monitoring and nutritional analysis. The project achieved impressive performance metrics with ~80% accuracy (mAP50), ~80% precision, and ~75% recall.

Key Features

  • Real-time food detection using YOLOv8
  • Support for 55 different food classes including fruits and vegetables
  • Calorie estimation per 100g of detected food items
  • Web interface with image upload and camera capture support
  • Bounding box visualization with confidence scores

Technical Details

Model Architecture & Implementation

FoodVision uses YOLOv8n (nano version) with the following specifications:

  • Input size: 640x640 pixels
  • Batch size: 32
  • Learning rate: 3e-4
  • Training epochs: 45

Food Categories

The system can detect various food items categorized by color:

  • Green foods: asparagus, avocados, broccoli, cabbage
  • White/Beige foods: banana, cauliflower, garlic, mushroom
  • Purple/Red foods: beetroot, blackberry, cherry, eggplant
  • Orange/Yellow foods: apricot, carrot, corn, mango

Team & Course Information

This project was developed by:

  • Brian Tham
  • Hong Ziyang
  • Javier Si Zhao Hong
  • Timothy Zoe Delaya

Let's get in touch

Contact me