Home PC News Carnegie Mellon’s robotic painter is a step toward AI that can learn...

Carnegie Mellon’s robotic painter is a step toward AI that can learn art techniques by watching people

Take the newest VB Survey to share how your organization is implementing AI in the present day.


Can a robotic painter be taught from observing a human artist’s brushstrokes? That’s the query Carnegie Mellon University researchers got down to reply in a study not too long ago revealed on the preprint server Arxiv.org. They report that 71% of individuals discovered the method the paper proposes efficiently captured traits of an authentic artist’s fashion, together with hand-brush motions, and that solely 40% of that very same group may discern the brushstrokes drawn by the robotic.

AI artwork era has been exhaustively explored. An annual worldwide competitors — RobotArt — duties contestants with designing artistically inclined AI techniques. Researchers on the University of Maryland and Adobe Research describe an algorithm referred to as LPaintB that may reproduce hand-painted canvases within the fashion of Leonardo da Vinci, Vincent van Gogh, and Johannes Vermeer. Nvidia’s GauGAN permits an artist to put out a primitive sketch that’s immediately reworked right into a photorealistic panorama by way of a generative adversarial AI system. And artists together with Cynthia Hua have tapped Google’s DeepDream to generate surrealist art work.

But the Carnegie Mellon researchers sought to develop a “style learner” mannequin by specializing in the strategies of brushstrokes as “intrinsic elements” of creative kinds. “Our primary contribution is to develop a method to generate brushstrokes that mimic an artist’s style,” they wrote. “These brushstrokes can be combined with a stroke-based renderer to form a stylizing method for robotic painting processes.”

Carnegie Mellon AI

Above: Brush strokes created by the generative mannequin.

The crew’s system includes a robotic arm, a renderer that converts photographs into strokes, and a generative mannequin to synthesize the brushstrokes based mostly on inputs from an artist. The arm holds a brush that it dips into buckets containing paints and places the comb to canvas, cleansing off the additional paint between strokes. The renderer makes use of reinforcement studying to be taught to generate a set of strokes based mostly on the canvas and a given picture, whereas the generative mannequin identifies the patterns of an artist’s brushstrokes and creates new ones accordingly.

To prepare the renderer and generative fashions, the researchers designed and 3D-printed a brush fixture geared up with reflective markers that may very well be tracked by a movement seize system. An artist used it to create 730 strokes of various lengths, thicknesses, and varieties on paper, which had been listed in grid-like sheets and paired with movement seize information.

In an experiment, the researchers had their robotic paint a picture of the fictional reporter Misun Lean. They then tasked 112 respondents unaware of the photographs’ authorship — 54 from Amazon Mechanical Turk and 58 college students at three universities — to find out whether or not a robotic or a human painted it. According to the outcomes, greater than half of the individuals couldn’t distinguish the robotic portray from an summary portray by a human.

Carnegie Mellon AI arm

Above: The robotic arm’s Misun Lean portray.

In the following stage of their analysis, the crew plans to enhance the generative mannequin by growing a stylizer mannequin that instantly generates brushstrokes within the fashion of artists. They additionally plan to design a pipeline to color stylized brushstrokes utilizing the robotic and enrich the educational dataset with the brand new samples. “We aim to investigate a potential ‘artist’s input vanishing phenomena,” the coauthors wrote. “If we keep feeding the system with generated motions without mixing them with the original human-generated motions, there would be a point that the human-style would vanish on behalf of a new generated-style. In a cascade of surrogacies, the influence of human agents vanishes gradually, and the affordances of machines may play a more influential role. Under this condition, we are interested in investigating to what extent the human agent’s authorship remains in the process.”

Most Popular

LinkedIn open-sources GDMix, a framework for training AI personalization models

LinkedIn recently open-sourced GDMix, a framework that makes training AI personalization models ostensibly more efficient and less time-consuming. The Microsoft-owned company says it’s an...

Google’s Smart Cleanup taps AI to streamline data entry

In June, Google unveiled Smart Cleanup, a Google Sheets feature that taps AI to learn patterns and autocomplete data while surfacing formatting suggestions. Now,...

ExamSoft’s remote bar exam sparks privacy and facial recognition concerns

Sometimes the light Kiana Caton is forced to use gives her a headache. On top of common concerns that come with taking a state...

Monetization strategies for the skyrocketing LatAm mobile game market (VB Live)

Presented by Facebook Audience Network There are hurdles to launching a mobile game internationally, but there’s never been a better time to go global. For...

Recent Comments