NVIDIA is doing some absolutely amazing stuff with its AI technology and GauGAN2 is just another notch in their belt in that area.
Using only raw text, NVIDIA’s latest project is capable of generating a beautifully rendered piece of art inspired by those words. DPReview cites examples such as “sunset on a beach” or “rocky beach” as prime iterations of what GauGAN2 is capable of making.
What apparently started out as a very rough project that generated less detailed output, GauGAN2 has since grown into a sophisticated program that illustrates the power of deep learning and AI.
As NVIDIA explains in their blog post, the GAN in GauGAN2 stands for generative adversarial networks and it is basically the magic behind how all of this works.
Ian Goodfellow of Google recently explained on AI Podcast how artificial intelligence uses generative adversarial networks to “learn.”
“If you take a deep neural network and you teach it to read, to actually look at photos and recognize letters that it can see in the photo, it can do that about as well as a human being can. But the process of learning to do it doesn’t look anything like the process that a human follows to learn to read.”
How it works is one network, called the generative network, produces output that is then evaluated by the discriminator network. In this case, the discriminator network evaluates the generative network’s output along the lines of whether the image created is authentic or not. If it isn’t, the discriminator network then produces feedback to guide the generative network’s next iteration. The two engage in this process until an accurate visual representation of the text in question is created.
“You can think of it as kind of like an art critic. The discriminator network looks at the image and says whether it’s real or fake. And it’s also able to tell the generator what it should do to make the image look slightly more realistic.”
Truly amazing stuff and capable of some amazing output that you can check out here.
We’d love to know your thoughts on GauGAN2 in the comments below.
Check out some of our other photography news on Light Stalking at this link right here.
[DPReview]