I keep getting criticisms about not matching ground truth in DeOldify. Yes. I get it. DeOldify didn't memorize every single thing in the world. Good luck making a model doing that. Now carry on while I make cool stuff.https://twitter.com/citnaj/status/1275884364996333570 …
-
That looks quite impressive to begin with. There will always be someone somewhere complaining that the 99% solution didn't solve the remaining 1% :D One question: do you have any blog post explaining the tech (at some details at least) behind the model?
There was this blog from last year: https://www.fast.ai/2019/05/03/decrappify/ … Of course that's quite bit dated now compared to what I'm working on currently. I don't even use a GAN anymore. But that's a good start.
-
-
You don't use a GAN anymore?
-
Haven't for a year actually. It's something else that's much more controllable based on the perceptual loss work talked about in FastAI on super resolution. There's quite a bit of modification to it to make it work.
End of conversation
New conversation -
-
-
Thanks for the link. Will check it out. Do you do all the work yourself (from research to model deployment) or do you work with a team?
Thanks. 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.