VitaNet

VitaNet is a radio frequency based contactless approach that accurately estimates the PPG signal using radar for stationary participants.

Dataset Setup

We have processed the data from 5 participants over 49 full nights preprocessed as Tensorflow Records (TFRs). Here are the steps to setup the dataset:

1- Download the TFRs from here. Please note that due to data limits for anonymous accounts, we have only uploaded a small subset of data. We will release the full dataset on publication of the paper.

2- Move the downloaded TFRs to code/tfrs.

Data is now set up!

Prerequisites

1- Install conda.

2- Install all the relevant packages through the .yml file that we have provided:

  cd code
  conda install -f vitanet.yml

Model Training

We have a provided a Makefile to run training and inference routines. For training:

  cd code
  make train

The default routine runs for 300 epochs, which may take a long time. To run a shorter training routine (50 epochs), run:

  make train.fast

Inference

For inference:

  make inference

Code Overview

common contains scripts to load data from TFRs.

code/models/paper.best contains the best VitaNet model that we were able to train.

code/tfrs contains the data from 5 participants over 49 full nights preprocessed as TFRs.

code/utils contains some utility functions.

code/vitanet.py contains all major model functionality.