I wonder if something like this could be combined with 3D animation to produce super realistic computer-animated films.
Nice idea.
Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
31–40 of 68 posts
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#32The progress in this field is astonishing. It wasn't even a few years ago that the typical demonstration in a paper on GAN would have been an array of very small images. I recall often seeing numbers between 64x64 and 256x256. At the time, the argument was raised that the resolution can hardly be increased because it is our brain that tries to identify objects and faces in the photos to figure out what they are repre…
> I know not every researcher or entrepreneur has the resources of NVIDIA to train for this many days, but let's not forget, that part needs to occur only once. It's not like they trained this on a GPU farm. According to the paper [1], they "trained the network on a single NVIDIA Tesla P100 GPU for 20 days". [1] http://research.nvidia.com/sites/default/files/pubs/2017-10_...
Yes, but they probably did not get it right on first try. You usually need hundreds of iterations to get it right. This does not imply 2000 days of GPU, but it quickly pushes this outside of reach of consumer hardware.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#33The progress in this field is astonishing. It wasn't even a few years ago that the typical demonstration in a paper on GAN would have been an array of very small images. I recall often seeing numbers between 64x64 and 256x256. At the time, the argument was raised that the resolution can hardly be increased because it is our brain that tries to identify objects and faces in the photos to figure out what they are repre…
> I know not every researcher or entrepreneur has the resources of NVIDIA to train for this many days, but let's not forget, that part needs to occur only once. It's not like they trained this on a GPU farm. According to the paper [1], they "trained the network on a single NVIDIA Tesla P100 GPU for 20 days". [1] http://research.nvidia.com/sites/default/files/pubs/2017-10_...
Where I live that thing alone is $17,000.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#34Worth pointing out the main criticism of GANs, which is that right now researchers don't really have a way to tell if a GAN is just copying and pasting the training data or not (there is no "test set" unlike in supervised learning). And in fact an ideal GAN could just learn to output the training set. One example someone found in the generated images for this model: https://twitter.com/nalkalchbrenner/status/92401333…
That is true, but if you don't look at them as a machine learning method, but rather as a computer graphics method, then it is quite impressive. It has the added benefit of being allowed to overfit as long as the average human does not find out. If you optimize for psychovisual metrics, GANs are fine.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#35Earlier quoted context omitted.
> I know not every researcher or entrepreneur has the resources of NVIDIA to train for this many days, but let's not forget, that part needs to occur only once. It's not like they trained this on a GPU farm. According to the paper [1], they "trained the network on a single NVIDIA Tesla P100 GPU for 20 days". [1] http://research.nvidia.com/sites/default/files/pubs/2017-10_...
Where I live that thing alone is $17,000.
P100 is essentially 1080Ti, so you can grab 1080 Ti and have the same speed. V100 might be the expensive version, up to 10x faster.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#36Earlier quoted context omitted.
Where I live that thing alone is $17,000.
P100 is essentially 1080Ti, so you can grab 1080 Ti and have the same speed. V100 might be the expensive version, up to 10x faster.
It may be the same hardware, but nVidia is known to block existing hardware features in their drivers on consumer level chips to drive prices. They did this in the past with quad buffer stereo (full screen only, not available through OpenGL) or decreased double precision performance (transparently added NOPs after each double precision opcode).
I will believe that these devices are equal when someone shows me benchmarks proving that. Until then, I am skeptical based on past experience.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#37Looking at synthetic celebrities is unsettling. They're all familiar. They're obviously celebrities, but it's impossible to remember their names.
Exactly as the actual celebrities then... I think i recognized one of them.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#38Well, I guess the propaganda bots will be a little more convincing now that they can have unique profile photos. Would be nice if I could use this to convince Facebook that some fictional image is myself, though.
Oh, damn. I've already had enough of fake personas, but now there is virtually no limit.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#39Worth pointing out the main criticism of GANs, which is that right now researchers don't really have a way to tell if a GAN is just copying and pasting the training data or not (there is no "test set" unlike in supervised learning). And in fact an ideal GAN could just learn to output the training set. One example someone found in the generated images for this model: https://twitter.com/nalkalchbrenner/status/92401333…
That is true, but if you don't look at them as a machine learning method, but rather as a computer graphics method, then it is quite impressive. It has the added benefit of being allowed to overfit as long as the average human does not find out. If you optimize for psychovisual metrics, GANs are fine.
But if you don't see it as a machine learning method, and don't care if the things the GAN spits out are just memorized photos, that means you don't actually care about the synthesis parts of the GAN? Thus the only reason to get excited about it is the interploration stuff; which significantly reduces how interesting it is in my eyes.
Re: Progressive Growing of GANs for Improved Quality, Stability, Variation [video]
#40Holy crap that video. It makes me think of the scramble suit from A Scanner Darkly[1] [1]: https://www.youtube.com/watch?v=BWne23FfKW8
Yea, but this looks much better.
http://blendberg.com/pl/aktualnosci