This month's Graphics & Rendering category includes 6 filed patent applications from 3 companies: EA (3), Sony (2), and Pixelworks Semiconductor Technology (Shanghai) Co.
(1). EA's filings cover a range of rendering techniques, from using neural networks and PCA codes to predict cloth mesh and normal maps based on character pose, to a per-pixel color blending approach for mesh particles, to a spectating system that streams simulated state packets to a client neural renderer. Sony's applications address AI-driven asset fidelity matching based on predicted in-game usage and a method for generating lightweight 3D placeholder representations of heavy assets to reduce cloud gaming client load. Pixelworks Semiconductor Technology (Shanghai) Co. filed a patent focused on filtering motion vector fields along their trajectory to reduce interpolation holes when generating in-between frames for smoother real-time rendering on handhelds and smartphones.
EA received 3 patents spanning several distinct rendering challenges. The first addresses cloth simulation by training a neural network on physics data to predict both a compressed mesh representation and a compressed normal map from a character's pose, then using a nearest-neighbor lookup against a pre-built texture database at runtime, replacing the need for per-frame physics calculations. The second covers a spectating system where simulated game state packets are streamed to a client that runs a machine learning model to render gameplay locally, allowing visual modifications and multiple viewpoints without requiring the full game application or its assets. The third takes a different direction entirely, describing a per-pixel color blending method for mesh particles in which each portion of a particle samples 1 of 2 colors, or a blend of both, based on its 3D distance to a defined point on a moving entity, producing effects that shift appearance based on spatial relationship rather than applying a single uniform color to the whole particle.
Sony received 2 patents, both oriented around reducing the computational cost of rendering in cloud and streaming contexts. One describes an AI model that takes player input and live game state as inputs, infers what purpose a given asset is serving at that moment, and uses that inferred context to set the fidelity at which the asset gets rendered and streamed, so rendering resources respond to how an asset is actually being used rather than its position on screen. The other applies a cloud-side pipeline that captures images of a large asset through a virtual camera, derives a compact 3D placeholder representation from those captures, and packages the result with metadata that tells the client rendering engine how to substitute it for the heavier original, reducing what the client must handle without requiring manual intervention.
Pixelworks Semiconductor Technology (Shanghai) Co. filed 1 patent aimed at frame interpolation for real-time 3D rendering on handhelds and smartphones. The application describes a method that filters a motion vector field specifically around disoccluded areas, areas where a moving foreground object uncovers background that had no prior motion vector coverage, by following the motion trajectory on a per-pixel basis rather than applying a uniform filter across the entire field. The result is a filtered motion vector field with smaller holes before the interpolated frame is produced, addressing a gap that simpler motion vector approaches leave when objects move across the frame.
All data sourced from USPTO patent filings. Google Patents may take several weeks to index recent publications. If a link is unavailable, search for the patent number at USPTO Patent Public Search.