To make a 3D model of an object, you generally need several cameras pointing at the object from different angles, images that are later processed to create the model in question.
Achieving it with a single camera from a single angle is not an easy task, but it is not impossible, and Facebook has proven it.
For the annual CVPR computer vision conference, Facebook is showing an algorithm that can generate a fairly detailed 3D model of a person wearing just a camera, something that can be very useful to continue creating projects for Oculus, its Virtual Reality platform. .
In a paper, called PIFuHD (already on github), three Facebook employees and a researcher at the University of Southern California propose a machine learning system to generate a high-detail 3D representation of a person and their clothing from a single 1K image. No depth sensor or motion capture equipment required.
It is the evolution of an older system they called PIFu, which could only handle relatively low resolution input images, limiting the precision and detail of the output model. PIFuHD now samples the input image to add fine surface details.
The objective of being able to do this is to create a realistic 3D avatar from the frontal capture of the user, which will help in the immersion of Virtual Reality projects.
Generating the avatar body is another step on the way to the ultimate goal set, allowing users to exist as their true physical selves in virtual environments, and see their friends as they really look, too.
There is still a long time for this technology to be available in the market, since Facebook itself says that they are years away for consumer products, such as videoconferencing with telepresence, for example.
