AI can now restore famous images using a 3D printer

Allows artificial intelligence and 3D printer to fight against the greatest threat to the art of the world?

It's such a team from MIT Computer Science and Artificial Intelligence Laboratory (CSAIL).

The new image reproduction design system employs 3D printing and deep learning to "authentically create favorite images" – regardless of lighting conditions or location.

"If you just reproduce the color of the picture as it looks in the gallery, it may look different in your house," said study co-author Changil Kim, a post-doctoral colleague at CSAIL.

So instead of using the traditional four solid inks (cyan, magenta, yellow, black) found in 2D printers, scientists used a special technique called "color-contoning."

The team trained a deep learning model that predicts an optimal supply of different inks (via MIT CSAIL)

The process includes a 3D printer and 10 different transparent inks that are stored in very thin layers – just like wafers and chocolate Kit-Kat bars.

By combining this method with a decade old halftone approach (which creates a gradient-like effect over dots), they were better able to capture the "color nuances".

Based on experimental replicas of various oil paintings created by a collaborator, the team found that RePaint was more than four times more accurate in producing accurate color shades than the most advanced physical models.

"Our system works under any lighting conditions, which shows much more color reproduction than almost any other previous work," Kim said.

There is one catch, though: CSAIL faksimile are just about the size of the business card. After all, 3D printing is not cheap.

In the future, however, they expect modern commercial printers to provide larger images and pave the way for a more efficient system.

Still, the question remains: What colors should be used for what images?

As with many tasks these days, people have gone over the burden of choice on a deeply learning model that can predict the optimal stack of colors in each area of ​​the canvas.

RePaint's creators represent their use in editing artwork for the home, protecting museum originals, and creating prints of historical pieces.

Researchers have found that RePaint is more than four times more accurate than the state-of-the-art physical models to create accurate color shades for various artworks (via MIT CSAIL)

The program has ways to go before "Star Night" duplicates.

For beginners, it can not completely reproduce certain colors, such as cobalt blue, due to the limited ink library that engineer Mike Foshey expects early.

And as you can see in the pictures and videos above, there is something that is clearly missing from RePaint's imitations: textures. The team will continue to work on better details or hoping to create special effects such as glossy and matte surfaces.

"The value of fine art has increased dramatically in recent years, so there is an increasing tendency for it to be closed in warehouses, away from the public eye," Foshey said. "We create technology to reverse this trend and create cheap and accurate reproductions that everyone can enjoy."

The complete report was published this week online by MIT CSAIL.

More coverage at

Source link