"deepart.io is the first online service which allows you to generate images styled to your favorite artist!
We apply an algorithm based on convolutional neural networks to combine the content of one image with the style of another image. The algorithm was introduced by Leon A. Gatys, Alexander S. Ecker, and Matthias Bethge in the research paper entitled 'A neural algorithm of artistic style'.
We are two researchers, trying to give the access to novel machine learning techniques to larger audience. Unfortunatelly the server costs a lot and we cannot afford many... Thus, if you like our work, please consider helping us with a small donation for covering server costs.
It's about art. We don't like ads (and they wouldn't cover expensive servers anyway). We also don't want to reduce the quality of the pictures. Your donation is our only hope :)
>We are two researchers, trying to give the access to novel machine learning techniques to larger audience. Unfortunatelly (sic) the server costs a lot and we cannot afford many... Thus, if you like our work, please consider helping us with a small donation for covering server costs.
Yep, I did read that right below the ETA line :-) still it's a long time, isn't it? It doesn't diminish the quality of the results anyway, I know it was a great effort to put everything together (and it is even greater to maintain). I just don't feel compelled to pay for it yet.
If you met the two guys who'd produced this, would you consider buying them a coffee? If so, the $1.99 to support some researchers and technology is hardly a big ask. You don't have to think of everything as 'paying for a good or service' - supporting folks doing something cool is just as valid (just don't tell the capitalist overlords).
How much GPU ram does the g2.8xlarge have? For printing an A4 at 300dpi you'd want 2480 x 3508px. If that doesn't fit in the GPU ram, and you are just doing this for yourself/friends it might be a better idea to run neural-style in CPU mode.
An order of magnitude slower but you'll have a lot more ram at your disposal
I was just writing the same comment, then said to myself: "hey don't wake up hackers, do it yourself :p" and i continue to my boring paid job like nothing happened.
The image I submitted is not the image shown to me on the submitted page (https://deepart.io/image/submissions/). Instead it appears to be someone else's.
Deep learning could become this decade's equivalent to the fractal craze of the 1990s.
It will have arrived when the next Jurassic Park remake-sequel features a Jeff Goldblum look-a-like who is an expert in "deep computing" and has lines like this:
"Of course the dinosaurs can rewrite computer programs! All this time, they've been watching, and their neural networks have been adapting... I told you it would happen."
This seems like a great application to use AWS Lambda. It would allow you to do things like charge PayPal for sub-minute results, or perhaps show multiple ads until you generate enough revenue to pay for the usage.
I had a similar project in progress, but paper author's university declined to licence technology. Did you somehow managed to convince them, or did you register as non-profit?
44 comments
[ 5.3 ms ] story [ 114 ms ] threadEDIT: Just noticed Latest submissions. Ignore this. :)
http://www.deepart.io/page/about/
"deepart.io is the first online service which allows you to generate images styled to your favorite artist!
We apply an algorithm based on convolutional neural networks to combine the content of one image with the style of another image. The algorithm was introduced by Leon A. Gatys, Alexander S. Ecker, and Matthias Bethge in the research paper entitled 'A neural algorithm of artistic style'.
We are two researchers, trying to give the access to novel machine learning techniques to larger audience. Unfortunatelly the server costs a lot and we cannot afford many... Thus, if you like our work, please consider helping us with a small donation for covering server costs.
It's about art. We don't like ads (and they wouldn't cover expensive servers anyway). We also don't want to reduce the quality of the pictures. Your donation is our only hope :)
Questions? E-mail us at admin@deepart.io
Łukasz Kidziński & Michał Warchoł"
It's going to be a long, long time before being able to comment on the results (of a photo I submitted), but the latest submissions look fantastic!
I don't want to wait forever for their service to generate pictures that are nowhere near printable :)
An order of magnitude slower but you'll have a lot more ram at your disposal
Did you try any of the other implementations of the paper?
(I know..., HN produced a lot of load on their servers...)
It will have arrived when the next Jurassic Park remake-sequel features a Jeff Goldblum look-a-like who is an expert in "deep computing" and has lines like this:
"Of course the dinosaurs can rewrite computer programs! All this time, they've been watching, and their neural networks have been adapting... I told you it would happen."
curl: (7) Failed to connect to deepart.io port 80: Connection refused
https://news.ycombinator.com/item?id=10526572