search

RameezI / dishscan

This repository explores AI/ML approaches towards automated tracking of dietary intake: estimating the ingredients & macro-nutrients in a food plate using digital images.
MIT License
2 stars 0 forks source link

readme

DISHSCAN

Welcome to the Dishscan Repository!

This project is dedicated to the exploration of Machine Learning (ML) approaches for automated analysis of a dish/food-plate for its ingredients and the nutritional content using machine vision.

We conducted a broad evaluations of ML vision models for food ingredient recognition, which are documented and discussed in this repository.

Comparison Table

About the Project

In the realm of dietary science and health, recognizing and understanding the nutritional content of food is crucial. This project focuses on developing and deploying machine learning models that can effectively identify and analyze the ingredients within a given dish or food-plate image. Our goal is to facilitate automatic understanding of the nutritional composition of various recipes and meals consumed in day-to-day life. This work focusses on evaluating the performance of various ML algorithms to detect ingredients or food-items present in a digital food image.
Recognition of the actual ingredients is one of the two essential ML functions towards our envisioned food assessment system depicted, the other is the portion estimation.

Publication

Our research in this field has been published in the paper titled "Food Ingredients Recognition Through Multi-label Learning" The publication can be accessed here.

[1] Ismail, Rameez, and Zhaorui Yuan, Food Ingredients Recognition Through Multi-label Learning. Embedded Artificial Intelligence: Devices, Embedded Systems, and Industrial Applications(2023): 105.

Getting Started

To get started with this project, follow these steps:

  1. Clone this repository to your local machine

    git clone git@github.com:RameezI/dishscan.git

  2. Create the dishscan conda environment: 

     conda env create -f environment.yml
 conda activate dishscan

  3. Make the dishscan package discoverable

    cd dishscan
pip install -e .

  4. Train a specific encoder/decoder setting

    python scanner.py --encoder MobileNetV1 --decoder GlobalAvgPoolDecoder

  5. To evaluate a trained model:

    # By default trained model is saved under a time_stamped `run_id` directory located at ~/tensorflow_logs/dishscan
# Assuming run_id directory is `20220813-0924` you can evaluate the saved model using the scanner script: 
python scanner.py --run_id 20220813-0924 --skip_training

For a complete list of options available to use with the scanner program:

 python scanner --help

Acknowledgments

The research leading to these results has received funding from the European Union's ECSEL Joint Undertaking under grant agreement number 826655 - project TEMPO. This activity was undertaken at Koninklijke Philips in partnership with the consortium mentioned. We appreciate the collective effort and collaboration that made this possible.

This project is maintained by: Rameez Ismail

Feel free to reach out for any questions, contributions, or collaboration opportunities.