Keras implementations of deep learning architectures used in our publication on heart rate estimation from BCG data.

The models are implemented with tf.keras of Tensorflow 1.13, but should work with most versions of Tensorflow.

You can install all required packages (except Tensorflow) with the provided requirements file:

git clone https://github.com/SamProell/bcg-hr-dl.git
cd bcg-hr-dl
pip install -r requirements.txt

The example jupyter notebooks highlight how to obtain and train the models. In essence, you can import any network from the models subfolder and use create to get a compiled Keras model:

# from models import stacked_cnn_rnn_improved as architecture
from models import baseline_fcn as architecture

patchsize, n_channels = 400, 1
model = architecture.create(input_shape=(patchsize, n_channels), enlarge=1)

# with x_data and y_data in the correct shape:
model.fit(x_data, y_data, batch_size=32)