A new system can faithfully reproduce complex artworks using new techniques in AI and in 3D printing.
There has been a lot of discussion about using AI to mimic conversation or images. At Springwise, we have covered innovations such as a social robot that uses conversational AI and AI-generated art. Now, a team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a new system which can reproduce paintings. Dubbed RePaint, it uses a combination of 3D-printing and deep learning.
While 2D printers commonly help in the reproduction of paintings, these use a fixed set of four inks. These colours normally include, cyan, magenta, yellow and black. The researchers at MIT found that they could capture a fuller spectrum of colour with a technique called ‘colour-contoning’. Using a 3D-printer to print from a palette of 10 different transparent inks stacked in very thin layers, colours appear more consistent. Additionally, the colours appear with virtually invisible spatial patterns on the surface. A deep-learning AI algorithm first analyses pictures of the original works to select the right combination of ink colours. This method is then combined with a second technique called half-toning, where an image is created by lots of little coloured dots rather than continuous tones. By combining these two techniques, the researchers were able to capture the nuances of the colours.
The result, according to the researchers, is a colour accuracy. This is up to four times greater than state-of-the-art physical models. Unlike standard prints, the colour of the reproduction is faithful to the original irrespective of placement or room lighting. Currently, the reproductions are only about the size of a business card. This is because of the expense and time it takes to complete the printing. In the future, the team hope to use commercial 3D printers to make larger paintings more efficiently.