Not bad, but as always currently
generated people seem to have 'random slightly out of focus patch syndrome' where there are patches of the images that are slightly out of focus (mainly on the face) and these don't match up with the physical reasons that cause objects in an image to be out of focus, those being more related to distances from the lens than difficulty of generation. Instead, it is hard to generate parts, like an eye (sometimes just the one), or the junction between a jumper and a neck where the jumper is out of focus but the skin right next to it is not.
Once you see it, you will always spot the current generation of generated faces. It is like someone's child played with the pictures and had a small amount of vaseline on their fingers then their parent tried to wipe it off and only got most of it.
(
My Name was saying these words inside your head on, Fri 15 Feb 2019, 10:01,
Share,
Reply)
Yeah, I noticed that too
(
mediocre ha ha ha, you're reading this, Fri 15 Feb 2019, 10:30,
Share,
Reply)
AIs never work the way even the people working on them think they work.
Humans have part of their sight-processing abilities hard wired; it's not all learned. Whereas 'learning' AIs often turn out to be focusing on completely unexpected factors. Like concentrating on textures rather than outlines in the visual field:
www.theregister.co.uk/2017/11/06/mit_fooling_ai/
(
joefish It's hard for thee to kick against the pricks, Fri 15 Feb 2019, 14:13,
Share,
Reply)