US 11,704,844 B2
View synthesis robust to unconstrained image data
Daniel Christopher Duckworth, Berlin (DE); Alexey Dosovitskiy, Berlin (DE); Ricardo Martin Brualla, Seattle, WA (US); Jonathan Tilton Barron, Alameda, CA (US); Noha Waheed Ahmed Radwan, Berlin (DE); and Seyed Mohammad Mehdi Sajjadi, Berlin (DE)
Assigned to GOOGLE LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Apr. 18, 2022, as Appl. No. 17/722,969.
Application 17/722,969 is a continuation of application No. 17/390,263, filed on Jul. 30, 2021, granted, now 11,308,659.
Claims priority of provisional application 63/059,322, filed on Jul. 31, 2020.
Prior Publication US 2022/0237834 A1, Jul. 28, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06T 11/00 (2006.01); G06T 7/90 (2017.01)
CPC G06T 11/001 (2013.01) [G06T 7/90 (2017.01); G06T 2207/20081 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computing system configured to perform view synthesis, the computing system comprising:
one or more processors;
a machine-learned view synthesis model configured to receive and process a position of a desired synthetic image to generate a synthetic image of a scene from the position,
wherein the machine-learned view synthesis model comprises a neural radiance field model,
wherein the machine-learned view synthesis model comprises a static content portion that models static content within the scene and a transient content portion that models transient occluders within the scene,
wherein the neural radiance field model has been trained on a set of training images that depict the scene, and
wherein the set of training images comprise unconstrained images that include the transient occluders; and
one or more non-transitory computer-readable media that collectively store instructions that, when executed by the one or more processors, cause the computing system to perform operations, the operations comprising:
obtaining the position of the desired synthetic image;
processing the position of the desired synthetic image with the machine-learned view synthesis model to generate the synthetic image; and
providing the synthetic image as an output.