New to computer vision and wondering which of the following categories are realistic to determine from photos? Photos will be similar to Facebook Profile photos. As I was working with Picassa 3 and it’s facial recognition, I was thinking about how it must categorize things and wondered what is possible now?

Some will be full body, some headshots. Assume multiple input photos to cross compare

Hair Color 
Hair Length 
Facial Hair 
Age 
Skin Tone (Black, Olive, Tan, White) 
Build (Skinny, Athletic, Average, Heavy, Obese) 
Face Shape (Heart, Square, Round, Diamond, Oval, Rectangular)

I would prefer to avoid API driven solutions like Azure. Too many experiences of developing something only to have the business model change. For environment I am leaning towards C#, but also have multiple high powered linux machines (dual 12 core, 512GB+ RAM…) available.

Feel free to add to the list of categories that can be determined.

What libraries would be used for which?