I generally added label error handling to the MMPDLoader.py file. Also fixed the associated YAML configs that use pre-trained models for testing - previously, the YAML config files used a pre-trained model file-path that would not work out-of-the-box for other toolbox users (including myself).
@McJackTang, my suspicion when you tested your previous PR (#134) is that you either used a different version of the MMPD dataset that you have locally and happens to have different motion labels, or perhaps there is some mistake in the labels in the MMPD dataset that was released and shared with dataset users via OneDrive and Baidu NetDisk. Either way, the error I ran into had to do with the motion labels for stationary motion not just being read as Stationary and Stationary (after exercise) as per the existing code in MMPDLoader.py, but also Watching Videos. My assumption is that Watching Videos also corresponds to a motion level of 1, or stationary motion type. Without the inclusion of Watching Videos as a potential label corresponding to the stationary motion type in this PR, users will encounter the same error I did and will be unable to use the dataset after pre-processing. The error ends up being encountered with this line when a motion label is not properly recognized and therefore an integer value is not present:
Fix Mini-MMPD motion label error, add label error handling, and fix YAMLs
I generally added label error handling to the
MMPDLoader.pyfile. Also fixed the associated YAML configs that use pre-trained models for testing - previously, the YAML config files used a pre-trained model file-path that would not work out-of-the-box for other toolbox users (including myself).@McJackTang, my suspicion when you tested your previous PR (#134) is that you either used a different version of the MMPD dataset that you have locally and happens to have different motion labels, or perhaps there is some mistake in the labels in the MMPD dataset that was released and shared with dataset users via OneDrive and Baidu NetDisk. Either way, the error I ran into had to do with the motion labels for stationary motion not just being read as
StationaryandStationary (after exercise)as per the existing code inMMPDLoader.py, but alsoWatching Videos. My assumption is thatWatching Videosalso corresponds to a motion level of 1, or stationary motion type. Without the inclusion ofWatching Videosas a potential label corresponding to the stationary motion type in this PR, users will encounter the same error I did and will be unable to use the dataset after pre-processing. The error ends up being encountered with this line when a motion label is not properly recognized and therefore an integer value is not present:https://github.com/ubicomplab/rPPG-Toolbox/blob/b483b40217edaefc00de595dea83a99b58517681/dataset/data_loader/MMPDLoader.py#L162