3D Subject Representations Using Layered Meshes for Efficient Rendering

The patent segments the complex volumetric representation into multiple 2D image layers captured from different perspectives. Instead of storing and processing a single large volumetric dataset, the system divides the 3D scene into multiple 2D projections, each representing a specific viewpoint. This segmentation reduces memory requirements while preserving rendering quality by allowing selective processing of relevant image layers based on the desired output perspective.

The patent transitions from working with pure 3D volumetric data to using 2D image layers with depth information. By representing the 3D scene as a stack of 2D images captured from different angles and depths, the system effectively uses another dimension (the layer index) to encode spatial information. This dimensionality change reduces computational complexity and memory usage compared to traditional volumetric rendering while maintaining the ability to generate high-quality 3D representations.

The patent performs preliminary action by capturing and preprocessing images from multiple perspectives before the actual rendering operation. The 3D scene is pre-sampled and organized into layered 2D image representations during an offline or preprocessing stage. This preliminary action stores the essential geometric and appearance information in a compact form, eliminating the need for intensive computational sampling during runtime rendering operations.

The patent creates 2D image layer copies of the 3D scene from different perspectives and depths. Instead of repeatedly sampling the original volumetric data during rendering, the system uses pre-generated 2D image copies that can be efficiently composited and processed. These image copies serve as lightweight proxies for the full volumetric data, maintaining modeling precision while dramatically reducing computational resource requirements during rendering operations.