New algorithmic system creates 3D animations with still photos

This system is seriously impressive

If you’ve ever wanted to bring a photo to life, researchers from the University of Washington may have a solution for you.

The researchers recently published a paper and YouTube video detailing their work on ‘Photo Wake-Up.’ The system can create a 3D animation from a single photo.

Not only are the animations impressively good, but you can also view them as video and interact with them through a monitor or in augmented or virtual reality.

Researchers included Chung-Yi Weng, a PhD student, Professor Brian Curless and Assistant Professor Ira Kemelmacher-Shlizerman.

The research team tested the system on 70 different photos downloaded from the internet. The test set included pictures of Stephen Curry, Goku and a Picasso painting.

Further, the team used a program called SMPL, which estimates 3D human poses and shapes from single images, along with deep learning to detect the photo’s subject and create a 2D cutout of it. Then, the program superimposed a 3D skeleton on it.

Additionally, the team used a warping algorithm to align the cutout and the skeleton.

Finally, the team used a proprietary user interface to correct any errors and help with texturing the animation. The interface ensured final animations were realistic and precise.

An algorithm also fills in the remaining space behind the moving subject of the image.

As cool as it is, the system isn’t perfect. It still can’t animate shadows or reflections. Also, the system may produce wrong animations due to “ambiguities.”

You can learn more about the Photo Wake-Up system through the research team’s website and research paper.

Via: Gizmodo