Here's a vision-based action detection repo: https://github.com/open-mmlab/mmaction2

I'm wondering if these deep learning based methods have been equivalent to those device based traditional methods?