A flat video call gives the remote person a rectangle. Their apparent position and scale belong to that screen, even if the screen is floating in a virtual room. A volumetric capture offers a different relationship: the viewer can move a little and see the person's shape change with the viewpoint.
The Mimesys reconstruction demonstration using four Intel RealSense cameras is a useful prompt for thinking about that difference. The camera count alone isn't the achievement. The important question is how separate partial observations become a coherent presence that can be rendered from another position.
A depth camera associates image locations with distance measurements. Given the camera model, those measurements can be turned into points in the camera's coordinate system. A second camera produces another set of points. To combine them, the system needs the position and orientation of each camera relative to a shared frame.
Multiple viewpoints help with visibility. If an arm hides part of the torso from one camera, another camera may see it. They don't guarantee complete coverage. A surface hidden from every camera remains unobserved, and a view from the opposite side of the capture setup may expose exactly those missing regions.
Intel's separate account of its four-camera Sundance capture makes the practical nature of that constraint clear. Camera placement, synchronization and occlusion all shape the usable result. Four cameras are a configuration to design around, not a universal specification for capturing an entire person from every possible angle.
Presence also depends on scale. Place a reconstructed person too small in a room and they become a figurine. Place them too large and the interaction feels physically strange even if the facial image is sharp. The receiving application needs a shared understanding of dimensions and floor position, not only a stream of colored points.
The social target changes which errors matter. A small geometric imperfection on a sleeve may be less disruptive than a delayed hand gesture or an unstable face. If a person points toward an object, the remote viewer needs the gesture and the object to agree in space and time. A high-resolution capture with inconsistent timing can make that relationship harder to follow.
I would therefore evaluate a system with ordinary conversational actions: turning toward another participant, indicating a location, moving a hand across the body. These reveal coverage and timing problems that a motionless portrait can hide.
The attraction is the possibility of sharing space rather than only sharing a picture. Achieving that doesn't require a perfectly reconstructed human at every instant. It does require the spatial cues that the interaction relies on to remain coherent. That is a more useful target than simply maximizing the number of points on screen.