FoodNet
FoodNet copied to clipboard
Cheng-K
Food Ingredients Recognition and Calorie Estimation using Multi-Task Deep Learning
Food Ingredients Recognition and Calorie Estimation using Multi-Task Deep Learning
View Demo
·
Report Bug
·
Request Feature
Table of Contents
-
About The Project
- Built With
-
Getting Started
- Prerequisites
- Installation
- Usage
- License
- Contact
- Acknowledgments
About The Project
This is my final year project for my undergraduate course. This project aims to propose a multi-task deep learning model that detects the category and ingredients of a captured food image and estimates the food calories and macronutrients consumed by users.
This project is made up of two repositories : this & FoodNet-App. This repository showcase the development of the deep learning model used in the project.
(back to top)
Built With
(back to top)
Getting Started
To get a local copy up and running follow these simple example steps.
Prerequisites
Installation
-
Open Anaconda Prompt
-
Clone the repo
git clone https://github.com/Cheng-K/FoodNet.git -
Change to project root directory
cd FoodNet -
Create new environment with environment.yml
conda env create -f environment.yml -
Activate the environment
conda activate PY38-TF28 -
Navigate to Food Datasets subdirectory and follow the instructions from each READMEs in each further subdirectory to set up the required datasets.
-
(Optional) Set up GPU by installing CUDA Toolkit 11.2 and cuDNN SDK 8.1.0 or follow the TensorFlow guide.
(back to top)
Usage
To effectively use this repository, the following shows the functionality of each directory and files.
Directory : Food Datasets
- Contains folders that store different types of dataset used in this project.
- A README is provided in each directory with attached sources and instructions to obtain and set up the dataset.
- Each folder also have a script that preprocess the stored dataset.
- The final-dataset stores the preprocessed dataset that is used for training the model after running all the preprocessing scripts for each dataset
Directory : Ingredient Embeddings
| File | Description |
|---|---|
| ingredient_embeddings_similarity.ipynb | Generates text embeddings to compare ingredients similarity so the missing nutritional information of an ingredient can be imputed with the most similar ingredient in the nutritional database. |
Directory : main
| File | Description |
|---|---|
| build_model.ipynb | Main file that contains code to build and train the proposed deep learning models |
| data_pipeline.ipynb | Prerequisite file for running build_model.ipynb. It serialize all the datasets into tfrecord files so that efficient data pipeline can be constructed to feed the data into the deep learning model |
In summary, the installation steps should set up everything including the datasets. All there is left is executing the build_model.ipynb to train the proposed deep learning model.
(back to top)
License
Distributed under the MIT License. See LICENSE.txt for more information.
(back to top)
Contact
Cheng Kei Ong - [email protected]
Project Link: https://github.com/Cheng-K/FoodNet
(back to top)
Acknowledgments
(back to top)
Cheng-K