HR-CNN

Spetlik, R., Franc, V., Cech, J. and Matas, J. (2018) Visual Heart Rate Estimation with Convolutional Neural Network. In Proceedings of British Machine Vision Conference, 2018

See http://cmp.felk.cvut.cz/~spetlrad/ecg-fitness/ for the original paper and the ECG-Fitness dataset.

This repo is being constructed. You can monitor the progress bellow.

Progress

• Document the installation.
• Convert models to a reasonable format.
• Create public evaluation scripts.
• Create public learning scripts.

Installation

We support only the following plug-and-play installation. You don't have to follow the steps bellow, but it may happen that it will not work :( This setup should work regardless of the operation system (i.e. Windows and Linux is OK).

1. Install miniconda http://lmgtfy.com/?q=miniconda+install.
2. Update miniconda with conda update -n base conda.
3. Create a Python 2.7 environment hr-cnn with conda create -n hr-cnn python=2.7.
4. Activate the environment with source activate hr-cnn.
5. Install packages.
   1. Install Pytorch http://lmgtfy.com/?q=install+pytorch.
   2. Run 'python -u install.py' OR perform the following procedure:
      1. Install docopt with conda install docopt.
      2. Install scipy with conda install scipy
      3. Install h5py with conda install h5py
      4. Install opencv with conda install opencv
      5. Install boost 1.65.1 with conda install boost=1.65.1
      6. Install bob.blitz with conda install bob.blitz
6. Clone the repo to a directory of your preference with git clone [email protected]:radimspetlik/hr-cnn.git.
7. Add the data.
   1. Download the models from http://cmp.felk.cvut.cz/~spetlrad/ecg-fitness/models.zip and extract them to data/models/.
   2. Copy the contents of the bbox directory (distributed in the 7zipped ECG Fitness database) to bob/db/ecg_fitness/data/bbox/.
   3. Copy the contents of the test_h5_faces.7zip to data\experiments\cnn\ecg-fitness-face-192x128\15\01\.
8. Run the test script with python hr-cnn/test.py. The network will evaluate a short sequence attached in the repo. You should get the following results:

[INFO] Root Mean Squared Error = ?
[INFO] Mean of error-rate percentage = ?
[INFO] Mean absolute error = ?
[INFO] Pearson's correlation = ?
[INFO] Pearson's correlation significance = ?