Curious if you were to compare this with standard webRtc and say Tensor Flow JS face landmark detection running against the mediastream, what is the difference?

Curious if you were to compare this with standard webRtc and say Tensor Flow JS face landmark detection running against the mediastream, what is the difference? edit: oh the "recognize your face" and compression I see (referring to Nvidia link someone else posted) wow