Some of the most famous paintings are so unique that it is impossible to recreate them accurately. While getting a basic reproduction is as simple as it can get, a 100%-accurate copy is hard to achieve. Nevertheless, researchers at MIT’s Computer Science & Artificial Intelligence Laboratory (CSAIL) successfully trained an artificial intelligence (AI) algorithm to recreate paintings using a 3D printer.
We have long seen 2D printers used to reproduce paintings. However, their capabilities are quite limited. They have a fixed set of only four colors, cyan, magenta, yellow and black, which makes it impossible to accurately mimic paintings. To overcome this limitations, CSAIL scientists created their RePaint system, which consists of an AI and AI-guided 3D printer. Their system uses a special painting technique they refer to as “color-contoning.”
Color-contoning involves using a 3D printer equipped with 10 different transparent inks placed in extremely thin layers to produce more accurate colors. MIT describes the thin layers as similar to the wafers and chocolate in a Kit-Kat bar. The team combined their color-contoning technique with an older technique called half-toning. Combining these two techniques, the team placed a lot of colored dots onto paintings instead of using continuous tones. This combined technique allowed the team to create more accurate colors and textures, which they then used to reproduce paintings using a 3D printer. The team also taught their AI to recognize colors and independently determine which ones are needed for each area of a painting.
A paper about the RePaint system will be presented at ACM SIGGRAPH Asia in December. Co-author Changil Kim explained how the system is different from other methods of reproducing paints.
“If you just reproduce the color of a painting as it looks in the gallery, it might look different in your home,” Kim said in a news release. “Our system works under any lighting condition, which shows a far greater color reproduction capability than almost any other previous work.”
The team members have been able to recreate several paintings using a 3D printer, although they will need a few more tweaks and improvements before taking on more complex masterpieces like Vincent van Gogh’s mesmerizing Starry Night. Mechanical engineer Mike Foshey, who’s also on the team, said they still can’t reproduce some colors due to some limitations of their ink library. They also haven’t been able to make their algorithm capable of predicting certain inks for some paintings. In the future, they also plan on adding more textures and reflections to the paintings to capture glossy and matte looks.
“The value of fine art has rapidly increased in recent years, so there’s an increased tendency for it to be locked up in warehouses away from the public eye,” Foshey said. “We’re building the technology to reverse this trend, and to create inexpensive and accurate reproductions that can be enjoyed by all.”
You can see how the AI recreates paintings using a 3D printer below: