Vincent AI Sketch Demo Draws In Throngs at GTC Europe
Cambridge Consultants showed off an deep-learning driven application this week at GTC Europe in Munich that lets you pick up a stylus and sketch out a few lines, and watch, in real time, as the application turns your squiggles into art in one of seven styles resembling everything from moody J.M.W Turner oil paintings to neon-hued pop art.
It’s a demo that stunned the more than 3,000 attendees during NVIDIA CEO Jensen Huang’s keynote speech Tuesday at the show.
Huang even climbed down from the stage to pick up a stylus and sketch a stylized NVIDIA logo and a profile of a man — which the application transformed into a Picasso-esque painting as he worked — grinning as the audience applauded.
The story behind the story: a finely tuned generative adversarial network that sampled 8,000 great works of art — a tiny sample size in the data-intensive world of deep learning — and in just 14 hours of training on an NVIDIA DGX system created an application that takes human input and turns it into something stunning.
Building on thousands of hours of research undertaken by Cambridge Consultants’ AI research lab, the Digital Greenhouse, a team of five built the Vincent demo in just two months.
The software is then loaded onto a PC or laptop running an NVIDIA GPU and paired with a Wacom tablet.
After Huang’s keynote, GTC attendees had the opportunity to pick up the stylus for themselves, selecting from one of seven different styles to sketch everything from portraits to landscapes to, of course, cats.
“It’s awesome,” said Christoph Angerer, a technologist, after seeing his quick sketch of a tree transformed into a work of abstract art.
The demo is radically different from two other deep learning demos that have intrigued the world over the past few years.
Google’s QuickDraw, for example, can quickly identify a quick sketch as stethoscope or a backpack.
A second type, style transfer, allows style of a great artist to be applied to a photograph or even a video.
But Vincent AI hints at a world — sketched out by Huang in his keynote — where humans will be able to set out a high-level direction for a new product design and rely on machines to fill in the details as they work.
The Vincent AI demo also speaks to both the power of DGX and generative adversarial networks, or GANs as they’re known.
While traditional deep learning algorithms have achieved stunning results by ingesting vast quantities of data, GANs create applications out of much smaller sample sizes by training one neural network to try to imitate the data they’re fed, and another to try to spot fakes.
The result, explained Monty Barlow, director of machine learning at Cambridge Consultants, is something his company can put to work on real-world problems, where a business might be dealing with a sample size of just a few hundred, rather than hundreds of thousands.
That’s a compelling story for the kinds of businesses Cambridge Consultants serves, who are looking for answers to particularly difficult problems.
But it’s also an application that’s captivated artists.
The app seems to sense where you’re going with a drawing, and flesh out your ideas as you work, in the style of a 17th century oil painting or a 20th century modernist sketch.
“Strangely, artists really like it,” Barlow said. “You learn about art, and what you know about art.”