Register now After registration you will be able to apply for this opportunity online.
Food Density Estimation with Machine Learning
Problem:
Accurately estimating the weight of food items is a significant challenge in healthcare applications. While state-of-the-art 3D cameras can precisely measure food volume, the lack of datasets with labeled food densities remains a major obstacle for accurately determining food amounts.
Goal of the thesis:
The thesis aims to create a dataset that includes the volume, weight, and 3D scans of various food items using a state-of-the-art structured light camera. Due to the vast variety of foods, compiling a comprehensive dataset is impractical. Therefore, the project will also include training and testing a machine learning model to predict the densities of food items that were not seen during its training.
Keywords: Food Science, Personalized Nutrition, Medical Nutrition, AI, Machine Learning, 3D Imaging, Food Tracking
Methodology
The methodology to tackle the challenge of estimating food densities in healthcare applications includes:
• Selecting a diverse list of food items to cook or purchase.
● Creating a food density dataset through conventional volume measurement methods and 3D imaging.
● Analyzing the dataset to identify clusters of foods with similar densities.
● Fine-tuning a deep neural network to predict food densities, adjusting the machine learning architecture as needed based on dataset analysis findings.
Startup Work Environment
We are a young and dynamic team of four ETH researchers who plan to found a spinoff this year. This is not just a thesis project; it's a step into an exciting start-up journey. If you share our passion for transforming healthcare, there may be an opportunity for you to join our team after completing your thesis.
Impact
Around 50% of patients in care facilities suffer from malnutrition. Our food tracking system automates nutrition tracking in these facilities by scanning the food when it is served and again upon collection, allowing us to accurately estimate what each patient has consumed. This automation enhances patient care through early detection and intervention for dietary deficiencies. In fact, early detection and treatment of malnutrition can reduce hospital stays by 30% and treatment complications by 20%, leading to potential health cost savings of 10 billion euros across Europe.
Methodology The methodology to tackle the challenge of estimating food densities in healthcare applications includes: • Selecting a diverse list of food items to cook or purchase. ● Creating a food density dataset through conventional volume measurement methods and 3D imaging. ● Analyzing the dataset to identify clusters of foods with similar densities. ● Fine-tuning a deep neural network to predict food densities, adjusting the machine learning architecture as needed based on dataset analysis findings.
Startup Work Environment We are a young and dynamic team of four ETH researchers who plan to found a spinoff this year. This is not just a thesis project; it's a step into an exciting start-up journey. If you share our passion for transforming healthcare, there may be an opportunity for you to join our team after completing your thesis.
Impact Around 50% of patients in care facilities suffer from malnutrition. Our food tracking system automates nutrition tracking in these facilities by scanning the food when it is served and again upon collection, allowing us to accurately estimate what each patient has consumed. This automation enhances patient care through early detection and intervention for dietary deficiencies. In fact, early detection and treatment of malnutrition can reduce hospital stays by 30% and treatment complications by 20%, leading to potential health cost savings of 10 billion euros across Europe.
The thesis aims to create a dataset that includes the volume, weight, and 3D scans of various food items using a state-of-the-art structured light camera. Due to the vast variety of foods, compiling a comprehensive dataset is impractical. Therefore, the project will also include training and testing a machine learning model to predict the densities of food items that were not seen during its training.
The thesis aims to create a dataset that includes the volume, weight, and 3D scans of various food items using a state-of-the-art structured light camera. Due to the vast variety of foods, compiling a comprehensive dataset is impractical. Therefore, the project will also include training and testing a machine learning model to predict the densities of food items that were not seen during its training.
Dr. Jotam Bergfreund (jotam.bergfreund@hest.ethz.ch),
and Dr. Raban Iten (itenr@ethz.ch),
Prof Dr. Peter Fischer,
Project page: www.alpinasana.ch
Dr. Jotam Bergfreund (jotam.bergfreund@hest.ethz.ch), and Dr. Raban Iten (itenr@ethz.ch), Prof Dr. Peter Fischer, Project page: www.alpinasana.ch