Outer Rim Archives
Archives · 2026 · 20260134607

Application (pre-grant publication)

TWO-DIMENSIONAL GAUSSIAN SPLATTING WITH TEXTURING FOR VIEW GENERATION IN COMPUTING SYSTEMS

Number
20260134607
Published
2026-05-14
Filed
2025-11-13
Assignee
DISNEY ENTERPRISES, INC.
Inventors
BRADLEY; Derek Edward, WEISS; Sebastian Klaus
CPC
G06T15/04; G06T15/06; G06T15/08; G06T15/20; G06T15/205
Verdict
Low Notable software
First reported
2026-W29 (2026-07-15)
Source
Google Patents · FreePatentsOnline

The keeper's note

Embodiments of the present disclosure provide generating one or more views of a scene from an input depicting the scene.

Abstract

Embodiments of the present disclosure provide generating one or more views of a scene from an input depicting the scene. An example method includes decomposing an input image depicting a view of a scene into a plurality of two-dimensional Gaussian primitives. Each two-dimensional Gaussian primitive may be defined based on a center position and a local tangent plane. A respective color grid is defined for each respective two-dimensional Gaussian primitive of the plurality of two-dimensional Gaussian primitives based on a parameterization of the local tangent plane. Colors for the respective color grid associated with each respective two-dimensional Gaussian primitive are interpolated based on a scaling factor, the interpolated colors corresponding to a texture associated with the respective color grid. One or more views of the scene other than the depicted view are constructed based on the interpolated colors for each respective color grid associated with each respective two-dimensional Gaussian primitive.

Background

BACKGROUND Field of the Various Embodiments

Embodiments of the present disclosure relate generally to computer vision and machine learning, and more specifically to techniques for texture-preserving view generation in computer vision systems. Description of the Related Art

Reconstructing three-dimensional scenes from a collection of two-dimensional images allows for various computer vision tasks to be performed. For example, these tasks may include view synthesis, in which unseen views are generated from one or more known views of a scene, relighting, animation playback, animation retargeting, and the like. Scene reconstruction can be performed from one or more input views of the scene using three-dimensional Gaussian splatting techniques. In three-dimensional Gaussian splatting techniques, a scene may be represented as a collection of volumetric primitives that are positioned and oriented in a three-dimensional space and then rendered as two-dimensional splats to generate a screen. To improve the rendering of details in a scene generated using Gaussian splatting techniques, two-dimensional Gaussian splatting techniques can be used. In two-dimensional Gaussian splatting techniques, the volumetric primitives of three-dimensional Gaussian splatting are replaced with two-dimensional primitives (or disks). However, these techniques generate views of scenes that may lack textural detail, especially in areas with limited geometric detailing.

Thus, what is neede

Claims

1. A processor-implemented method, comprising: decomposing an input image depicting a view of a scene into a plurality of two-dimensional Gaussian primitives, each two-dimensional Gaussian primitive being defined based on a center position and a local tangent plane; defining a respective color grid based on a parameterization of the local tangent plane for each respective two-dimensional Gaussian primitive of the plurality of two-dimensional Gaussian primitives; interpolating colors for the respective color grid associated with each respective two-dimensional Gaussian primitive based on a scaling factor, the interpolated colors corresponding to a texture associated with the respective color grid; and constructing one or more views of the scene other than the depicted view based on the interpolated colors for each respective color grid associated with each respective two-dimensional Gaussian primitive. || 8. A processing system, comprising: at least one memory having executable instructions stored thereon; and one or more processors configured to execute the executable instructions to cause the processing system to: decompose an input image depicting a view of a scene into a plurality of two-dimensional Gaussian primitives, each two-dimensional Gaussian primitive being defined based on a center position and a local tangent plane; define a respective color grid based on a parameterization of the local tangent plane for each respective two-dimensional Gaussian primitive of the plurality of two-dimensional Gaussian primitives; interpolate colors for the respective color grid associated with each respective two-dimensional Gaussian primitive based on a scaling factor, the interpolated colors corresponding to a texture associated with the respective color grid; and construct one or more views of the scene other than the depicted view based on the interpolated colors for each respective color grid associated with each respective two-dimensional Gaussian primitive. || 15. A non-transitory computer readable medium having executable instructions stored thereon which, when executed by one or more processors, performs an operation comprising: decomposing an input image depicting a view of a scene into a plurality of two-dimensional Gaussian primitives, each two-dimensional Gaussian primitive being defined based on a center position and a local tangent plane; defining a respective color grid based on a parameterization of the local tangent plane for each respective two-dimensional Gaussian primitive of the plurality of two-dimensional Gaussian primitives; interpolating colors for the respective color grid associated with each respective two-dimensional Gaussian primitive based on a scaling factor, the interpolated colors corresponding to a texture associated with the respective color grid; and constructing one or more views of the scene other than the depicted view based on the interpolated colors for each respective color grid associated with each respective two-dimensional Gaussian primitive.