新型人体影像捕捉可用于任何场景画面

Capturing human performances in high-definition* 3-D is a complicated process and among the many challenges is getting the lighting right.

“The Relightables,” an impressive new project from Google researchers, puts the subject in the center of an LED* installation and the resulting 3-D models are remarkable — and more importantly, relightable.

Ordinary volumetric* capture technology uses multiple cameras in a 360-degree setup to capture what can look like a photorealistic* representation of a subject, including all the little details like clothing deformation*, hair movement, and so on.

It has two serious weaknesses: First, it’s more like a 3-D movie than a model, since you can’t pose the person or change their attributes* or clothing; the second is an extension of the first, in that you can’t change the way the person is lit — whatever lighting they had when you captured them, that’s what you get.

“The Relightables” is an attempt by a team at Google AI to address this second issue, since the first is pretty much baked in. Their system not only produces a highly detailed 3-D model of a person in motion, but allows that model to be lit realistically by virtual* light sources, making it possible to place it in games, movies, and other situations where lighting can change.

The installation is lined with 331 LED lights that can produce any color, and as the person is being captured, those LEDs shift in a special structured pattern that produces a model. The resulting models can be placed in any virtual environment.

The examples aren’t exactly Hollywood-level quality, but being relightable brings these performances a lot closer to ordinary 3-D models than they were before. Of course, you still have to do all your acting inside the installation.

(SD-Agencies)