Identifying humans in a video

Questions
, , ,
1

I’m looking for some good methods for finding humans in a video, I’m using

Entity["Concept", "Person::93r37"]

with limited accuracy.

(1) Is there a better entity I should be using?
(2) Is there a good training set that I could use to improve performance? At this point, I don’t care what they are doing, I would just like to identify that there is a person in the frame.

Any advice would be appreciated.

2

I think I now know the answer to the first part of this one. Because Mathematica’s high level machine vision uses MXNet, it’s limited on how MXNet performs and the standard training available.

I have since been able to use some demo code for YOLO which seems to work better.

In relation to the training data, any sources of additional good training data would be really appreciated.