StyleGAN2 • Hacker News Board

Re: StyleGAN2

#31

Post by Veedrac » Fri, Dec 13, 2019, 6:09 PM UTC

nuccy wrote: post #28

ImminentFate wrote: post #21

Earlier quoted context omitted.

On your site I can consistently get 100% by looking at the backgrounds since they generate in somewhat inorganic patterns.

Also after watching the video from the StyleGAN2 team https://drive.google.com/file/d/1f_gbKW6FUUHKkUxciJ_lQx29mCq... now I know that original StyleGAN, images from which are apparently used for this "game", produces faces with a "water droplet" and phase artifacts, so I was able to spot few fakes just by looking for those things.

Only whichfaceisreal.com uses the original StyleGAN. The github.io links use StyleGAN2.

Re: StyleGAN2

#32

Post by Veedrac » Fri, Dec 13, 2019, 6:19 PM UTC

TomVDB wrote: post #30

ImminentFate wrote: post #21

Earlier quoted context omitted.

On your site I can consistently get 100% by looking at the backgrounds since they generate in somewhat inorganic patterns.

When not the ears, it's the background. If the real pictures had the background removed, I'd have a very hard time scoring 100%.

That's the point of the hard version.

Re: StyleGAN2

#33

Post by globuous » Fri, Dec 13, 2019, 7:26 PM UTC

resiros wrote: post #2

The demo in the official video is mind blowing. https://www.youtube.com/watch?v=c-NJtV9Jvp0 I wonder when we will see full movies unrecognizable from real ones made from deep learning.

Insane the part where they get multiple angles from the same generated face

Re: StyleGAN2

#34

Post by gdubs » Fri, Dec 13, 2019, 7:52 PM UTC

Of course we’ll hit a wall at some point, but when this repo dropped the other night and I saw the rotating faces in the video, it made me realize that in the future, VR experiences might be generated with nets rather than modeled with traditional CG.

Re: StyleGAN2

#35

Post by narsk » Fri, Dec 13, 2019, 8:12 PM UTC

ctfu at the car images. I made a twitter bot to tweet them out with fake make/model names https://twitter.com/bustletonauto

Re: StyleGAN2

#36

Post by TeeWEE » Thu, Dec 19, 2019, 3:19 PM UTC

TeeWEE wrote: post #7

gwd wrote: post #3

Only watched the video, but one of the interesting things is the potential method to tell a generated image from a real one: namely, if you take a generated image, it's possible to find parameters which will generate exactly the same image. But if you take a real image, it's generally not possible to get exactly the same image, but only a similar one. The exact point in the video: https://youtu.be/c-NJtV9Jvp0?t=208

This is only possible if you have access to the model neural net... If you dont you cant tell the difference.

Actually i was wrong. Generative Adverserial Nets often work accross Machine Learning models...