Re:This Artwork Does Not Exist
by MarkusWandel on Wednesday March 04, 2020 @06:48PM
(#22489439)
Funny that of all the "This X does not exist" sites, this one only gives relatively small, low-res pictures. Because I'd totally print out some of that stuff and hang it up!
[ Parent
| on HN ]
Re:This Artwork Does Not Exist
by valentinvieriu on Thursday March 05, 2020 @01:25AM
(#22491393)
I've launched a similar project some weeks ago. https://art42.net see the previous thread https://news.ycombinator.com/item?id=22344254
Hopefully you all don’t mind me tagging on (again) with a link to my project — similar space, but with Fauvist art:
http://gregorywieber.com/art/a-walk-through-latent-space-mak...
Some of those pictures are actually quite good.Would it be technically possible to generate larger versions?
Re:This Artwork Does Not Exist
by himinlomax on Thursday March 05, 2020 @08:28AM
(#22493321)
Note http://www.art42.fr/
But then it would exist :)
This generator of all of them seems to raise the more existential questions. What does it mean to exist? Surely art is art regardless of whether it's on a physical canvas. Just the act of generating it and serving pixels changes the state from "doesn't exist" to "does exist"...
Mount a flat screen on the wall and refresh periodically. I would do that if it could be trained with specific artists I like.
It's exponentially more expensive to train these at higher resolutions.
Not exponentially. More like linearly in pixel count, you just tack on another convolution block. (StyleGAN doesn't use self-attention, but even naive self-attention is just quadratic.)
The problem is more that you hit diminishing returns fast from training at higher resolutions. There's not too much difference between training at 1024px or training at 512px and using an off-the-shelf superresolution NN upscaler, but the latter is like 4x faster. So why bother? You don't always even have input data which is 1024px+.