Here's a problem:
Although machine learning and artificial intelligence have almost approached human-level performance at recognizing and identifying objects in images, there is a gap in a computer's ability to produce images.
Here's a possible solution:
When a painter gazes upon a beautiful landscape, he can understand the trees, rivers, and clouds to recreate the scene on his canvas. Just like the painter who can reproduce what he recognizes using a series of strokes on canvas, computers can learn to create images via a series of strokes.
Autodesk Research is part of the Office of the CTO that creates stories about the future and helps Autodesk make them come true. Kevin Frans is an intern working on Reinforcement Learning for Interactive Design. Chin-Yi Cheng is a Senior Research Scientist. Kevin and Chin-Yi are working on the solution.
Check out Kevin and Chin-Yi's interactive blog post where you can play with the knobs and see the effects:
If you want to fully understand the science rather than just play with the knobs, check out their research paper:
Autodesk has always been an automation company, and today more than ever that means helping people make more things, better things, with less; more and better in terms of increasing efficiency, performance, quality, and innovation; less in terms of time, resources, and negative impacts (e.g., social, environmental). Encoding images as a series of high-level constructs, such as brush strokes or discrete shapes, can often be key to both human and machine understanding. This is one element in the future of making.
Painting is alive in the lab.