Localization of Activity

Hi,

thanks for this contribution.

My question is - how to localize the activity in a scene? e.g. Two people are dancing and I want to find the coordinates of these people. My understanding is that currently it may not be possible with this architecture as the Object detection -> Object interaction is not explicitly learned here.

If that is the case, do you have any suggestions to extend to that level?

Thanks a tonne.

Arun