1/ Neural Networks like to cheat a lot. Especially when it comes to something as ambiguous and hard to train for as the arbitrary colors of clothing. So default "safe bets" are what the models tend to do as a result. You'd think GAN training would be a good way to circumvent
-
-
2/ this but I'm finding out that I can get the same results without GAN training now (shhh, it's a secret lol). So the next question is- was I just using GAN training ineffectively......? That's where I'm at now.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
It would be nice- though I don’t know how practical- to be able to provide hints to the network like - “you may want to make this gray, but I know it’s red”
-
There’s something called exemplar based models that allow you to do exactly that. Trick is how to marry the current functionality with that. I have ideas :)
End of conversation
New conversation -
-
-
It is the GAN's weakness in global encoding of relationships, different colours should show different tonal curves across a form that has both direct (warm to neutral) and indirect (cool) lighting, but the GAN is not able to know this, or so it would seem.
#GANflawsThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.