Watch the Mona Lisa turn her head, in today's edition of stupid StyleGAN encoder tricks... #StyleGAN github.com/pbaylies/stylegan… cc: @aarlo @Zergfriend @neurokinetikz @calamardh @roadrunning01 @Miles_Brundage

4:21 PM · Jun 5, 2019

18
310
31
982
31,805
...enhance! (using L-BFGS this time, and some other refinements...)
2
0
2
9
1,085
Wow. Adding to my list to check out! My first silly idea is: if you use a non-human StyleGAN model, but ask it to find the latent representation of a picture of Donald Trump, would you get the piece of furniture (or whatever model) that most resembles Donald Trump?
2
0
0
2
There's a good paper about this; you can find decent approximations in the latent space, but it will be meaningless to the model, i.e. interpolation will be garbage
1
0
0
2
This is interesting, however, the real mona lisa definitely has a chubbier face.
0
0
0
3
Pretty nice but seems in between the ends it's a bit different person.
1
0
0
3
I think that's a bit of an optical illusion, but I know what you mean! Here's another comparison... (artist's rendering vs. encoding)
4
0
0
2
I wonder how this would compare to traditional image morphing (e.g. psychomorph)
1
0
0
2
It does a great job of image morphing too!
2
0
0
9
448