During 2010-2012, I was part of a startup called Clownfish Media. We basically created something very similar to this and got scary accurate results then. Given how accessible computer vision has become, the image in the tweet comes at no surprise to me.

Best part - we got a first gen raspberry pi to crunch all the data locally at 2-5fps. Gender, age group (child, youth, teen, young adult, middle age, senior), and approximate ethnicity were all recorded and logged. Everyone had a unique profile and could track people between cameras and days (underlying facial features do not change).

Next time you look at digital signage, just be aware that it is probably looking back at you.