MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202621057754 A) filed by Indian Institute Of Technology Gandhinagar on May 06, 2026, for Method And System For Incremental Multi-Scene Neural Radiance Field Representation.
Inventors include Singh, Prajwal; Tiwari, Ashish; Vashishtha, Gautam; and Raman, Shanmuganathan.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: ABSTRACT METHOD AND SYSTEM FOR INCREMENTAL MULTI-SCENE NEURAL RADIANCE FIELD REPRESENTATION The present disclosure relates to a computer-implemented system (100) for incremental multi-scene neural radiance field representation. The encoding module (104) encodes spatial coordinates and viewing directions using multi-resolution hash encoding to generate feature representations. The scene conditioning module (106) assigns pseudo-labels and generates conditioned feature representations. The neural rendering model (102), comprising multi-layer perceptrons (128, 130), predicts density (122) and color (126) for a plurality of scenes within a shared parameter space. The training module (108) incrementally trains the neural rendering model (102) using synthetic views of previously learned scenes without storing original training data. The rendering module (110), based on camera parameters and pseudo-labels, performs volumetric integration to generate rendered outputs on an output device (116).
Disclaimer: Curated by HT Syndication.
