View Proposal

---

Proposer

John See

Title

FOODIE: Deployment of Models for Food Image Understanding

Goal

Development of a suite of API endpoints for AI models to analyse and interpret food visual data, and a simple demonstration of its usages

Description

Artificial Intelligence (AI) models for food images use computer vision and deep learning techniques to analyse, recognise, and interpret food-related visual data. Many of these models can classify dishes, estimate portion sizes, identify ingredients, assess food quality, and even predict calorie content from images. This project aims to develop a suite of SaaS (Software-as-a-Service) RESTful API endpoints for AI models that extract information from food images. Some feasible services to start with include food image classification, food image segmentation and food calorie estimation. The implemented APIs can be demonstrated via a simple program (web or mobile) that takes in an input image (or a set of images from a gallery) and returns the results to be displayed and visualised at the front-end. *As some available models may be dependent on GPU for inference, a feasible solution should be engineered - either by cloud hosting with compute (costly), or by finding alternative lightweight models for local hosting.

Resources

Possible packages to use: https://github.com/LARC-CMU-SMU/FoodSeg103-Benchmark-v1; https://github.com/KennyYao2001/16824-CaLORAify

Background

Reasonable level of competency in programming, especially Python; Some familiarity with deployment of machine learning models and API programming would be an added advantage

Url

Difficulty Level

Moderate

Ethical Approval

None

Number Of Students

1

Supervisor

John See

Keywords

artificial intelligence, model deployment, restful api

Degrees

Bachelor of Science in Computing Science