README in progress, for now please refer to

• [email protected]
• GitHub main project
• Child Growth Monitor Website

Child Growth Monitor Machine Learning

Introduction

This project uses machine learning to identify malnutrition from 3D scans of children under 5 years of age. This one-minute video explains.

Getting started

Requirements

Training the models realistically requires using GPU computation. Project members are currently using a variety of cloud instances (GCS, AWS, Azure) and local machines for training. A separate backend project is currently developing the DevOps infrastructure to simplify this.

You will need:

• Python 3
• TensorFlow GPU
• Keras

Installation

These steps provide an example installation on a local Ubuntu workstation from scratch:

• Install Ubuntu Desktop 18.04.1 LTS
• Install NVIDIA drivers
  Please note that after rebooting, the secure boot process will prompt you to authorize the driver to use the hardware via a MOK Management screen.

sudo add-apt-repository ppa:graphics-drivers
sudo apt-get update
sudo apt-get install nvidia-390
sudo reboot now

• Install Anaconda with Python 3.6

conda update anaconda
conda update python
conda update --all
conda create --name cgm
source activate cgm
conda install tensorflow-gpu
conda install ipykernel
conda install keras
conda install vtk progressbar2 glob2 numbs pandas
pip install --upgrade pip
pip install git+https://github.com/daavoo/pyntcloud

Dataset access

Data access is provided on as-needed basis following signature of the Welthungerhilfe Data Privacy & Commitment to Maintain Data Secrecy Agreement. If you need data access (e.g. to train your machine learning models), please contact Markus Matiaschek for details.

Preparing training data

• Run python create_pickle.py
• Run the notebook create_dataset.ipynb

Training the models

• Run the notebook train_on_dataset.ipynb

This currently takes around 6 hours to run on a local NVIDIA GTX 1080 Ti.

Evaluating the results

TODO

Contributing

Please see CONTRIBUTING.md for details.

Versioning

Our releases use semantic versioning. You can find a chronologically ordered list of notable changes in CHANGELOG.md.

License

This project is licensed under the GNU General Public License v3.0. See LICENSE for details and refer to NOTICE for additional licensing notes and use of third-party components.