Really impressive stuff. Could this same technique be used on human photos to transfer traits like gender, age, ethnicity?
Could be really useful for those age-progressed photos used in missing persons listings, for instance.
Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
91–100 of 146 posts
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#92I've only recently started reading about deep neural networks, and the thing that strikes me the most about the literature is the lack of mathematics . Open a NIPS paper from 2010 or so, and you'll see extremely dense mathematics: nonparametrics, variational approximation, sampling theory, riemannian geometry. But from my (admittedly small) sampling of the convnet / RNN literature there really doesn't seem to be much…
Wait until we have neural networks designing neural networks.
Been there, done that. Beats humans :)
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#93Earlier quoted context omitted.
The next step would be to have it actually turned into an actual canvas rendering, with the texture and such of actual paint. Could probably be done using 3d printer technology, inkjet, and/or robotics in some manner... ...in fact, I know you can already get photos printed to canvas - but taking it to the next stage of texture would be amazing - right now, I think the best you can get is to have a trained person "hig…
What for? Really, honest question. If you look at it as an achievement of technology, then it's fine. But if you look at it as real thing from real person, then it's fake.
I prefer the aesthetic qualities of paint applied by strokes over an inkjet print of a "painterly" image. I'd enjoy a service that allows you to upload source images, pick a stylization, and then buy a painted-by-robotically-wielded-brush version.
By way of analogy: I own some furniture from Ikea and some nice antique furniture. I don't know or care much about the provenance of the antique furniture. It's made up of bargains from yard/estate sales. If a big box store could sell me an inexpensive antique-alike bureau that was indistinguishable from my existing one (to unaided human senses), I'd happily buy it. I want the thing more than the story behind the thing.
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#94It wasn't highlighted on the github readme, but I think that the satellite photo to map and map to satellite photo (!!!) is incredible as well: https://taesung89.github.io/cyclegan/2017/03/25/maps-compari...
i don't even know how that's possible. How could it recover a golf course?
Did it recover a golf course? If you're talking about examples #1 and #5, the golf course is the input, and the output is a map.
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#95I've only recently started reading about deep neural networks, and the thing that strikes me the most about the literature is the lack of mathematics . Open a NIPS paper from 2010 or so, and you'll see extremely dense mathematics: nonparametrics, variational approximation, sampling theory, riemannian geometry. But from my (admittedly small) sampling of the convnet / RNN literature there really doesn't seem to be much…
Too bad that Math heavy approaches just cannot compete.
We have a winner.
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#96it feels like there's a new deep learning paper each week, ever so slightly bringing me closer to an existential nervous breakdown.
Yeah, I feel that. I tell myself they're only tools, really no stranger than time frequency domain transformations to someone unfamiliar with the fourier transform. Yet there really is something disturbing about seeing a computer resurrect so much of the mind of an artist who's been dead for nearly a century. I know the GPU doesn't understand what it's doing, but did Monet? When I paint I don't really understand. The…
I think that is only true on first glance. But if you look a little deeper, none of the photo->painting are really accurate. The painters had more than color scheme as information. Van Gough, for example, did not always use a correct perspective. In general, the deeper meaning of art is not in the technique but in the perspective of the artist.
That said, I am amazed by the results even so.
I also want to note here, that I don't see any reason why machine intelligence could not produce meaningful works of art but it will require a new way of looking at it.
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#97I've only recently started reading about deep neural networks, and the thing that strikes me the most about the literature is the lack of mathematics . Open a NIPS paper from 2010 or so, and you'll see extremely dense mathematics: nonparametrics, variational approximation, sampling theory, riemannian geometry. But from my (admittedly small) sampling of the convnet / RNN literature there really doesn't seem to be much…
Wait until we have neural networks designing neural networks.
Neural Architecture Search with Reinforcement Learning
https://arxiv.org/abs/1611.01578
Too bad, it is already there :P
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#98Earlier quoted context omitted.
Too bad that Math heavy approaches just cannot compete.
We have a winner.
Yes we do. In terms of spending time mathurbating what is fundamentally limited and incapable, I suggest we spend more time play with the shiny new toys that do fullfill its very own promises.
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#99I've only recently started reading about deep neural networks, and the thing that strikes me the most about the literature is the lack of mathematics . Open a NIPS paper from 2010 or so, and you'll see extremely dense mathematics: nonparametrics, variational approximation, sampling theory, riemannian geometry. But from my (admittedly small) sampling of the convnet / RNN literature there really doesn't seem to be much…
So what we're seeing is these fields pre-deep learning were mathematical disciplines making steady progress on well-understood foundations. Then deep learning came in, was exceptionally effective at problems that had been difficult to crack, and people shifted focus because it seems weird to be diddling around with incremental gains on techniques that are significantly less effective. What this created though, was a…
> The theory will eventually catch up
Maybe... but the emergent behavior in a complicated system (and these networks are only getting more complicated, not less) is likely to quickly become more complex than the human mind can reasonably be expected to understand, given any amount of time.
We actually know a lot less about biology, for example, than your typical "biology 101" course would lead you to believe. It's pretty cool that we figured out how DNA translates into proteins, but actually figuring out how a cell decides which DNA to express at any given time, much less what any arbitrary protein actually does, takes you right back into guess-and-check territory. There are heuristics you can use to eliminate some guesses more quickly than others, but we're so far away from a unified explanatory theory of biology that I'm doubtful one will ever be developed.
Re: Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
#100Earlier quoted context omitted.
So what we're seeing is these fields pre-deep learning were mathematical disciplines making steady progress on well-understood foundations. Then deep learning came in, was exceptionally effective at problems that had been difficult to crack, and people shifted focus because it seems weird to be diddling around with incremental gains on techniques that are significantly less effective. What this created though, was a…
Agreed with the first paragraph, but I think this sort of behavior is normal for most disciplines. That is, it's not CS looking like psychology or sociology, it's normal for CS to do this. I think we're seeing a similar thing right now in biology with CRISPR.
FWIW, CRISPR isn't a great analogy. It's a well understood mechanism, and not really all that different from standard genome editing techniques like restriction enzymes, zinc-finger nucleases, etc. It's new and flashy and super powerful -- and the applications of CRISPR are still being discovered -- but the mechanism is understood and simple.