iBVP dataset, iBVPNet model and MACC evaluation metrics

[jnj256]

Dear rPPG-Toolbox team,

Below we summarize our contributions that have been published in our recent work on iBVP dataset.

Through our work, we primarily contribute a novel rPPG dataset, which we refer to as 'iBVP dataset'. Additionally, we introduce iBVPNet model, which was found to be an effective 3D-CNN architecture, with results presented in the paper. We further discover the use of maximum amplitude of cross-correlation (MACC) as highly suitable evaluation metrics to validate rPPG methods.

Researchers will be able to access our dataset by referring to the details mentioned on this GitHub page.

Our code builds upon the rPPG-Toolbox repo, and all the contributions from our side can be found in this forked repo. We would like to create a pull request to merge our contributions. For the same, we summarize the key changes below, and provide a link for detailed difference log.

Added:

• dataset/data_loader/IBVPLoader.py: For loading RGB video frames of iBVP dataset
• neural_methods/model/iBVPNet.py: The iBVPNet model introduced in our work
• neural_methods/trainer/iBVPNetTrainer.py: Trainer for iBVPNet
• "train" and "infer" configs for iBVP dataset as well as iBVPNet model

Updated:

• dataset/data_loader/init.py: Added import statement for IBVPLoader
• neural_methods/trainer/init.py: Added import statement for IBVPNet
• evaluation/metrics.py: Updated for MACC evaluation metrics
• evaluation/post_process.py: Updated for MACC evaluation metrics
• unsupervised_methods/unsupervised_predictor.py: Updated for MACC evaluation metrics
• main.py: Added support for iBVPNet model
• config.py: Updated for iBVPNet model
• setup.sh: Updated to make the installation work with Ubuntu 20.04 (can be ignored while merging)
• ReadMe: Added details of the iBVP dataset and iBVPNet model

[yahskapar]

I will get to testing and reviewing this PR fully by this weekend @jnj256 - sorry for the wait!

[jnj256]

I will get to testing and reviewing this PR fully by this weekend @jnj256 - sorry for the wait!

Thanks Akshay for the update. I identified few refinements for which I shall raise a separate PR later, as these should not affect your testing. For e.g., at line 139 of metrics.py, there should be no unit for MACC ("dB" needs to be removed).

[jnj256]

@yahskapar : Thank you for going through all the changes made and for your suggestions. I've now addressed your comments and suggestions. Please have a look if everything looks good to go.

[yahskapar]

@jnj256, looks like there's some conflicts, probably because of the OMIT PR that was merged recently. Can you pull those changes and make sure to resolve any conflicts? You should be good to squash and merge after that - thanks again for your contribution!

[jnj256]

@yahskapar : Thanks for your quick review of the changes. About the conflict related to OMIT PR, I've now resolved it.