IDP in Informatics: Deep Dish - Real-Time Food Identification and Pricing Solution for Large-Scale Dining Facilities

05.11.2024, Diplomarbeiten, Bachelor- und Masterarbeiten

The project focuses on developing an AI-powered food detection system for large-scale cafeterias to automate food recognition, optimize pricing, and improve efficiency. Using pretrained models and public datasets, it aims to evaluate and train a model for high-accuracy food detection, streamlining operations in commercial food settings.

Abstract
This IDP aims to develop and evaluate a machine learning-based food detection system tailored for high-volume environments like canteens and cafeterias. The primary objective is to create an automated, efficient, and scalable solution for food recognition, which can be applied to optimize service processes, streamline pricing, and improve operational speed. The project will leverage pretrained models and publicly available datasets, assessing and fine-tuning them to meet the accuracy and efficiency demands of real-world implementations.

Background
In high-traffic food service environments, automation can significantly enhance the efficiency and accuracy of operations. Traditional food identification and pricing methods can be slow, inconsistent, and reliant on manual labor, making them suboptimal for large-scale implementations. This project proposes the development of a computer vision-based food detection algorithm capable of identifying various food items on a tray, enabling rapid and automatic food recognition. Such a solution would streamline the pricing process, improve throughput, and potentially reduce human error in food service environments.
The project will capitalize on pretrained models and publicly available datasets, ensuring an effective implementation.

Tasks
• Model Evaluation and Baseline Development: Implement and compare two different pretrained models to establish a baseline for food detection accuracy and speed in high-volume settings.
• Training with Public and Custom Data: Train a selected model using publicly available food recognition datasets and, if necessary, augment with proprietary data to enhance model accuracy and adaptability to specific environments.

Prerequisites
• Proficiency in deep learning techniques, particularly for image analysis.

References
• https://www.kaggle.com/datasets/sainikhileshreddy/food-recognition-2022• https://paperswithcode.com/datasets?task=food-recognition
• Rokhva, S., Teimourpour, B., & Soltani, A.H. (2024). Computer Vision in the Food Industry: Accurate, Real-time, and Automatic Food Recognition with Pretrained MobileNetV2. arXiv preprint arXiv:2405.11621.

To apply, please send a short mail with CV and transcript of records to florian.hinterwimmer@tum.de.

Kontakt: florian.hinterwimmer@tum.de