Researchers with the University of Washington and Facebook have detailed a method using artificial intelligence to animate a person using a single still image. The algorithm is called Photo Wake-Up, and it will be presented at the Conference on Computer Vision and Pattern Recognition on June 19.

The Photo Wake-Up algorithm is given a single still image, such as a photo of a person standing or even an image of a less-than-realistic painting. The system animates the character or person featured in the still image, enabling it to step out of the photo and move forward in 3D space. The hole in the image where the character was located is automatically filled in by the software.

According to the study, the method can create a 3D character from the still image that is capable of walking, running, sitting, and jumping in 3D. The resulting animations can be experienced using augmented reality, enabling artwork in museums to literally walk off the wall, for example.

Despite the input image only providing a single camera position, the resulting 3D model can be viewed from the side and back, as well. The quality varies based on the image; a sample video shared by the researchers shows some 3D models that look more realistic than others.

As has already been demonstrated with AI-based faced generation technologies, it's likely the quality of this method will improve greatly over coming months and years. The study follows a different method revealed by Samsung in May that can transform a still image of a face into an animated, talking video.

2019-6-20 16:52