Vincent AI, Refined through Generative Adversarial Network

Nvidia held a public demonstration of "Vincent AI," a deep learning-based application developed by Cambridge Consultants, at the GTC Europe event held in Munich, Germany.

The Vincent application allows users to complete simple sketches drawn with a stylus in real-time as artwork in seven different styles, including oil paintings by J.M.W. Turner, a 19th-century English landscape painter, and neon-colored pop art.

The public demonstration, which took place during Nvidia founder and CEO Jensen Huang's keynote address, drew gasps of amazement from over 3,000 attendees. CEO Jensen Huang came down from the stage during the keynote and personally drew the Nvidia logo and a human face with a stylus. When the sketch was transformed in real-time into a Picasso-style painting, the audience burst into applause.

Behind the remarkable performance of Vincent AI lies a GAN (Generative Adversarial Network) that has undergone precise tuning. This application, created by sampling 8,000 artworks and undergoing 14 hours of training on Nvidia's DGX system, transforms information input by humans into fantastic works of art.

Based on research conducted over thousands of hours by Cambridge Consultants' AI research lab, the Digital Greenhouse, a team of five was able to build the Vincent demo in just two months. The software can be installed on a PC or notebook equipped with an Nvidia GPU and used with a Wacom tablet.

Following CEO Jensen Huang's keynote address, GTC participants were also given the opportunity to pick up a stylus themselves, choose one of seven styles, and try drawing portraits, landscapes, cats and other subjects to experience the application.

This demonstration represents a fundamentally different technology from two deep learning demos that have captured global attention in recent years. For example, Google's QuickDraw tends to misidentify simple sketches as stethoscopes or bags. Another type, style transfer, applies a specific painter's style to photographs or videos. In contrast, Vincent AI enables the possibility of having humans set high-level directions during new product design processes while allowing the machine to fill in the remaining details.

Through this Vincent AI demonstration, the computing power of DGX and the remarkable performance of GAN have also been proven. While conventional deep learning algorithms achieved impressive results by processing vast amounts of data, GAN enables the creation of applications with far fewer samples than before by training one neural network to mimic input data and another neural network to identify fabricated portions.

Monty Barlow, Director of Machine Learning at Cambridge Consultants, stated that this achievement can be applied to solving real-world problems. He noted that companies can leverage it even when sample sizes number only in the hundreds rather than hundreds of thousands. This is generating considerable expectations among companies seeking Cambridge Consultants' consulting services to solve challenging problems.

Director Barlow stated, "Surprisingly, many artists have shown considerable interest in this application," and added, "Using Vincent AI will allow you to understand what art is and how well-versed you are in knowledge about art."