Graphics & Rendering

Nvidia received 1 patent describing a deep neural network system that generates multiple interpolated frames for each rendered frame in real-time gaming and VR applications. The technology differs from existing frame generation methods, which typically produce only a single interpolated frame between rendered frames, by extending the approach to create several generated frames per original frame. The system maintains the low-latency performance required for competitive gaming and immersive VR experiences while boosting perceived frame rates without adding performance overhead.

Intel's single patent filing covers a generative AI system that transforms the visual style of classic games during active gameplay. The approach applies style-altering generative models dynamically to frames or assets as they render, enabling aesthetic transformations that go beyond resolution upscaling. This real-time processing eliminates the need for offline asset preparation or storing large volumes of pre-enhanced content, allowing visual remixing to happen on-the-fly as players experience the game.

Ubisoft Entertainment filed 1 patent for a texture compression technique that represents high-resolution physically-based rendering materials as neural networks stored in GPU memory. Each material is encoded as a compact neural network trained to perform both decompression and trilinear filtering in fewer than 200 GPU operations, with network weights stored as 16-bit floating-point values. The system applies block compression techniques to the neural feature representations themselves and uses end-to-end training with perceptual loss functions to learn and counteract compression artifacts, a capability that fixed mathematical compression methods cannot replicate.