You can move your mouse anywhere on the screen and get different images for each point. Moving over the the cat vs. the fluid simulator doesn't make a difference, except that the points are mirrored over the Y-axis.
Yup that is correct! I do like the idea of the cats looking at the mouse cursor. Would need to do some work figuring out how to interpolate just the pose part of the latent space.
People with views like these shouldn't be given oxygen with platforms that allow them to form community around horrible and hateful ideas. In my opinion they should feel left out, until they realize that it is their immoral views that make them unacceptable to larger society.
Unbelievable that they managed to make the same mistake twice. Especially when the first mistake was so bad that neighbors complained enough that it had to be paved over.
Anyone who watches any content creator at all will notice that they are already paranoid wrecks due to existing copyright law. Seems like once or twice an hour a streamer will wonder out loud if they've inadvertently DMCAed themselves, even while carefully trying to avoid doing so.
With this law, the risk is going to be so high that even carefully treading won't be worth it. It will completely put the industry on ice.
What worries me even more than information that has been exfiltrated is what backdoors have been left behind. Even when the hatch is closed, how can you be totally sure that there is nothing left behind on the accessed systems?
Could someone who is more familiar with security comment on this?
But at first I thought this was a serious app to be used for recruiting. I actually thought it was a clever way to screen candidates quickly for their coding ability. I was a little disappointed that it wasn't that.
My jaw has been frequently misplaced recently by the amount of lies that people seem to believe. But I think it became permanently detached when finding out at the very end that people _actually_ believed them.
I initially built this just as a fun landing page for a larger AI related project. I ended up building a 2D interpolator for GAN generated outputs. Since the control was 2D, I figured it would be fun to add a fluid simulator that tracks your mouse. And now it's become a fun interactive toy all on its own.
The way it works is by taking a 3x3 grid of GAN outputs, applying a gaussian kernel for blending that maximizes total variance, interpolating the frames and then finally producing a video file. As you move your mouse, it seeks to various points within the video.
I have started with cats here because all great things on the internet lead back to cats, of course. But this can work with any GAN that can be interpolated. In this case, it is Stylegan2.
There are some easter eggs in here as well - see if you can find them :)
Feedback is definitely appreciated. Thank you for taking a look.