A great example here:
https://msit.microsoftstream.com/video/040716a6-44f1-420e-b642-29f562b60bd9?channelId=fa304e27-4640-...
- There is one woman falsely labeled for much of the video
- Many other people are labeled, but really there should be a threshold to recognize that people aren't very important in this video (there aren't, for example, a few key speakers) and so the People module really is pretty noisy and the labeling feels pretty low quality / irrelevant.