A version of the paper here https://songma.github.io/files/hm_video.pdf

I really question this data sample.