LogIn
The research team presented a new generative adversarial network called Alias-Free GAN that does not depend on absolute pixel coordinates.
One of the problems with most GANs is that despite their hierarchical convolutional nature, the synthesis process of typical models depends on absolute pixel coordinates leading to glued artifacts. You've probably seen these generated faces that don't look real enough in motion due to some details being stuck at certain points.
"We trace the root cause to careless signal processing that causes aliasing in the generator network," wrote NVIDIA in a new paper. "Interpreting all signals in the network as continuous, we derive generally applicable, small architectural changes that guarantee that unwanted information cannot leak into the hierarchical synthesis process."
The research discusses a new network that generates natural-looking results better suited for video and animation. NVIDIA also shared some videos showing much more smooth results but the code is not available yet. You can learn more here.
You can also check out a great breakdown by Two Minute Papers:
Don't forget to join our new Reddit page, our new Telegram channel, our Discord, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.