Instead of searching for the ideal model for a photo shoot, photographers of the future may be able to generate one using artificial intelligence. Neural networks these days can generate portraits of imaginary people.
NVIDIA recently published a paper titled “Progressive Growing of GANs for Improved Quality, Stability, and Variation.” GAN stands for “generative adversarial network,” and it’s a system that uses two neural networks — one generates things and the other evaluates them. These algorithms are capable of generating artificial “photos” that look real to humans.
For its project, NVIDIA found that training the neural network using low-resolution photos of real celebrities and then ramping up to high-res photos helped to both speed up and stabilize the “learning” process, allowing the AI to create “images of unprecedented quality.”
To demonstrate the system, NVIDIA trained the neural network using the CelebA HQ database of famous faces.
Once it had been trained on these real faces, the AI was able to begin generating photo-realistic photos of fake people.
Here’s a 6-minute video that shows more examples of what this neural network can do:
In addition to faces, the neural network can also generate “photos” of objects and scenes.
“While the quality of our results is generally high compared to earlier work on GANs, and the training is stable in large resolutions, there is a long way to true photorealism,” the researchers conclude in their paper. “That said, we feel that convincing realism may now be within reach.”
(via NVIDIA via Engadget)