Piecewise-polynomial coupling layers for warp-predicting neural networks

Number: 10818080
Published: 2020-10-27
Filed: 2018-10-11
Assignee: Disney Enterprises, Inc.
Inventors: Muller; Thomas, McWilliams; Brian, Rousselle; Fabrice Pierre Armand, Novak; Jan
CPC: G06N3/09; G06N3/0475; G06N3/08; G06N7/01; G06T15/506; G06N3/045; G06T15/005; G06T15/06; G06N3/10; G06N3/0499
Verdict: Low Notable software
Source: Google Patents · FreePatentsOnline

The keeper's note

Neural-network warp-prediction rendering technique (granted).

Abstract

According to one implementation, a system includes a computing platform having a hardware processor and a system memory storing a software code including multiple artificial neural networks (ANNs). The hardware processor executes the software code to partition a multi-dimensional input vector into a first vector data and a second vector data, and to transform the second vector data using a first piecewise-polynomial transformation parameterized by one of the ANNs, based on the first vector data, to produce a transformed second vector data. The hardware processor further executes the software code to transform the first vector data using a second piecewise-polynomial transformation parameterized by another of the ANNs, based on the transformed second vector data, to produce a transformed first vector data, and to determine a multi-dimensional output vector based on an output from the plurality of ANNs.

Background

BACKGROUND(1) When rendering images with path-tracing algorithms, light paths need to be constructed to connect emitters to sensors. The sampling distributions used for constructing these paths directly influence the estimation error, i.e., noise, and the efficiency of rendering. A large body of research has been devoted to developing methods for constructing high-energy light paths, such as bidirectional path tracing, metropolis light transport, or offsetting inefficiencies by reusing computation (e.g. photon mapping, many-light rendering, gradient-domain rendering, and/or control variates). While these algorithms perform well in certain applications, they tend to under-perform in others.SUMMARY(2) There are provided systems including warp-predicting neural networks having piecewise-polynomial coupling layers, and methods for use by such systems, substantially as shown in and/or described in connection with at least one of the figures, and as set forth more completely in the claims.

Claims

1. A system comprising: a computing platform including a hardware processor and a system memory storing a software code including a plurality of artificial neural networks (ANNs); the hardware processor configured to execute the software code to: partition a multi-dimensional input vector into a first vector data and a second vector data; transform the second vector data using a first piecewise-polynomial transformation parameterized by a first ANN of the plurality of ANNs, based on the first vector data, to produce a transformed second vector data; transform the first vector data using a second piecewise-polynomial transformation parameterized by a second ANN of the plurality of ANNs, based on the transformed second vector data, to produce a transformed first vector data; and determine a multi-dimensional output vector based on an output from the plurality of ANNs, wherein the output is produced using the transformed first vector data from the second ANN and the transformed second vector data from the first ANN. 11. A method for use by a system including a computing platform having a hardware processor and a system memory storing a software code including a plurality of artificial neural networks (ANNs), the method comprising: partitioning, using the hardware processor, a multi-dimensional input vector into a first vector data and a second vector data; transforming, using the hardware processor, the second vector data using a first piecewise-polynomial transformation parameterized by a first ANN of the plurality of ANNs, based on the first vector data, to produce a transformed second vector data; transforming, using the hardware processor, the first vector data using a second piecewise-polynomial transformation parameterized by a second ANN of the plurality of ANNs, based on the transformed second vector data, to produce a transformed first vector data; and determining, using the hardware processor, a multi-dimensional output vector based on an output from the plurality of ANNs, wherein the output is produced using the transformed first vector data from the second ANN and the transformed second vector data from the first ANN.