公开(公告)号：US20230083929A1

公开(公告)日：2023-03-16

申请号：US17984132

申请日：2022-11-09

申请人： Pixar ,  Disney Enterprises, Inc.

发明人： Thijs Vogels ,  Fabrice Rousselle ,  Jan Novak ,  Brian McWilliams ,  Mark Meyer ,  Alex Harvill

IPC分类号： G06T5/00 ,  G06N3/08 ,  G06T15/06 ,  G06T15/50 ,  G06N5/04 ,  G06N20/00 ,  G06F17/18 ,  G06N3/04 ,  G06T5/50

摘要： A modular architecture is provided for denoising Monte Carlo renderings using neural networks. The temporal approach extracts and combines feature representations from neighboring frames rather than building a temporal context using recurrent connections. A multiscale architecture includes separate single-frame or temporal denoising modules for individual scales, and one or more scale compositor neural networks configured to adaptively blend individual scales. An error-predicting module is configured to produce adaptive sampling maps for a renderer to achieve more uniform residual noise distribution. An asymmetric loss function may be used for training the neural networks, which can provide control over the variance-bias trade-off during denoising.

公开(公告)号：US11532073B2

公开(公告)日：2022-12-20

申请号：US16050314

申请日：2018-07-31

申请人： Pixar ,  Disney Enterprises, Inc.

发明人： Thijs Vogels ,  Fabrice Rousselle ,  Jan Novak ,  Brian McWilliams ,  Mark Meyer ,  Alex Harvill

IPC分类号： G06T15/50 ,  G06T15/06 ,  G06T7/90 ,  G06T5/00 ,  G06T7/00 ,  G06N3/04 ,  G06N3/08 ,  G06N5/04 ,  G06N7/00 ,  G06N20/00 ,  G06F17/18 ,  G06T5/50

摘要： A modular architecture is provided for denoising Monte Carlo renderings using neural networks. The temporal approach extracts and combines feature representations from neighboring frames rather than building a temporal context using recurrent connections. A multiscale architecture includes separate single-frame or temporal denoising modules for individual scales, and one or more scale compositor neural networks configured to adaptively blend individual scales. An error-predicting module is configured to produce adaptive sampling maps for a renderer to achieve more uniform residual noise distribution. An asymmetric loss function may be used for training the neural networks, which can provide control over the variance-bias trade-off during denoising.

公开(公告)号：US11037274B2

公开(公告)日：2021-06-15

申请号：US16789025

申请日：2020-02-12

申请人： PIXAR ,  Disney Enterprises, Inc.

发明人： Thijs Vogels ,  Fabrice Rousselle ,  Brian McWilliams ,  Mark Meyer ,  Jan Novak

IPC分类号： G06K9/62 ,  G06T5/00 ,  G06T5/50 ,  G06N3/08 ,  G06N3/04 ,  G06K9/46 ,  G06T7/00 ,  G06T15/06 ,  G06T7/90

摘要： Supervised machine learning using neural networks is applied to denoising images rendered by MC path tracing. Specialization of neural networks may be achieved by using a modular design that allows reusing trained components in different networks and facilitates easy debugging and incremental building of complex structures. Specialization may also be achieved by using progressive neural networks. In some embodiments, training of a neural-network based denoiser may use importance sampling, where more challenging patches or patches including areas of particular interests within a training dataset are selected with higher probabilities than others. In some other embodiments, generative adversarial networks (GANs) may be used for training a machine-learning based denoiser as an alternative to using pre-defined loss functions.

公开(公告)号：US10818059B1

公开(公告)日：2020-10-27

申请号：US16514852

申请日：2019-07-17

申请人： Pixar

发明人： Fernando Ferrari De Goes ,  Alonso Martinez ,  Michael B. Comet ,  Patrick Coleman

IPC分类号： G06T13/20 ,  G06T17/20 ,  G06T13/40

摘要： Embodiments provide for sculpt transfer. Embodiments include identifying a source polygon of a source mesh that corresponds to a target polygon of a target mesh. Embodiments include determining a first matrix defining a first rotation that aligns a target rest state of the target polygon to a source rest state of the source polygon, determining a second matrix defining a linear transformation that aligns the source rest state to a source pose of the source polygon, wherein the linear transformation comprises rotating and stretching, determining a third matrix defining a second rotation that aligns the source pose to the target rest state, and determining a fourth matrix defining a third rotation that aligns the source rest state to the source pose. Embodiments include determining a target pose of the target polygon based on the target rest state, the first matrix, the second matrix, the third matrix, and the fourth matrix.

公开(公告)号：US10789686B2

公开(公告)日：2020-09-29

申请号：US16735079

申请日：2020-01-06

申请人： PIXAR ,  Disney Enterprises, Inc.

发明人： Thijs Vogels ,  Fabrice Rousselle ,  Brian McWilliams ,  Mark Meyer ,  Jan Novak

IPC分类号： G06K9/62 ,  G06T5/00 ,  G06T5/50 ,  G06N3/08 ,  G06N3/04 ,  G06K9/46 ,  G06T7/00 ,  G06T15/06 ,  G06T7/90

摘要： Supervised machine learning using neural networks is applied to denoising images rendered by MC path tracing. Specialization of neural networks may be achieved by using a modular design that allows reusing trained components in different networks and facilitates easy debugging and incremental building of complex structures. Specialization may also be achieved by using progressive neural networks. In some embodiments, training of a neural-network based denoiser may use importance sampling, where more challenging patches or patches including areas of particular interests within a training dataset are selected with higher probabilities than others. In some other embodiments, generative adversarial networks (GANs) may be used for training a machine-learning based denoiser as an alternative to using pre-defined loss functions.

公开(公告)号：US10366184B1

公开(公告)日：2019-07-30

申请号：US15941928

申请日：2018-03-30

申请人： Pixar

发明人： Theodore W. Kim ,  Fernando Ferrari De Goes ,  Breannan D. Smith

IPC分类号： G06T17/20 ,  G06F17/50 ,  G06T15/08

摘要： Systems, methods and articles of manufacture for rendering images depicting materials are disclosed. A stable Neo-Hookean energy model is disclosed which does not include terms that can produce singularities, or require the use of arbitrarily selected clamping parameters. The stable Neo-Hookean energy may include a length-preserving term and volume-preserving term(s), and the volume-preserving terms themselves may include term(s) from a Taylor expansion of a logarithm of a measurement of volume. The stable Neo-Hookean energy may further include an origin barrier term that increases the difficulty of reaching the origin and expands a mesh in response to a perturbation when the mesh is at the origin. Closed-form expressions of eigenvalues and eigenvectors of a Hessian of the stable Neo-Hookean energy are disclosed, which may be used in a simulation of a material to, e.g., project the Hessian to semi-positive-definiteness in Newton iterations used to determine a substantially minimal energy configuration.

公开(公告)号：US10319133B1

公开(公告)日：2019-06-11

申请号：US13295094

申请日：2011-11-13

申请人： Kurt Fleischer ,  Warren Trezevant ,  Andrew Witkin

发明人： Kurt Fleischer ,  Warren Trezevant ,  Andrew Witkin

IPC分类号： G06T13/00 ,  G06T13/40

摘要： Users may dynamically specify a “posing root” node in an animation hierarchy that is different than the model root node used to define the animation hierarchy. When a posing root node is specified, users specify the pose, including translations and rotations, of other nodes relative to the posing root node, rather than the model root node. Poses of nodes may be specified using animation variable values relative to the posing root node. Animation variable values specified relative to the posing root node are dynamically converted to equivalent animation variable values relative to the model root node, which then may be used to pose an associated model. Animation data may be presented to users relative to the current posing root node. If a posing root node is changed to a different location, the animation data is converted so that it is expressed relative to the new posing root node.

公开(公告)号：US10282885B2

公开(公告)日：2019-05-07

申请号：US15858206

申请日：2017-12-29

申请人： PIXAR

发明人： Alexis Angelidis

IPC分类号： G06T13/60 ,  G06T15/08 ,  G06T15/20 ,  G06T15/50 ,  G06T17/00

摘要： A multi-scale method is provided for computer graphic simulation of incompressible gases in three-dimensions with resolution variation suitable for perspective cameras and regions of importance. The dynamics is derived from the vorticity equation. Lagrangian particles are created, modified and deleted in a manner that handles advection with buoyancy and viscosity. Boundaries and deformable object collisions are modeled with the source and doublet panel method. The acceleration structure is based on the fast multipole method (FMM), but with a varying size to account for non-uniform sampling.

公开(公告)号：US20180293713A1

公开(公告)日：2018-10-11

申请号：US15946654

申请日：2018-04-05

申请人： PIXAR

发明人： Thijs Vogels ,  Fabrice Rousselle ,  Brian McWilliams ,  Mark Meyer ,  Jan Novak

IPC分类号： G06T5/00 ,  G06T5/50 ,  G06N3/08 ,  G06N3/04

CPC分类号： G06T5/002 ,  G06K9/4628 ,  G06K9/623 ,  G06K9/6257 ,  G06K9/627 ,  G06K9/6298 ,  G06N3/04 ,  G06N3/0454 ,  G06N3/0472 ,  G06N3/08 ,  G06N3/084 ,  G06T5/50 ,  G06T7/0002 ,  G06T7/90 ,  G06T15/06 ,  G06T2207/20076 ,  G06T2207/20081 ,  G06T2207/20084 ,  G06T2207/20192 ,  G06T2207/30168 ,  G06T2207/30201

摘要： Supervised machine learning using neural networks is applied to denoising images rendered by MC path tracing. Specialization of neural networks may be achieved by using a modular design that allows reusing trained components in different networks and facilitates easy debugging and incremental building of complex structures. Specialization may also be achieved by using progressive neural networks. In some embodiments, training of a neural-network based denoiser may use importance sampling, where more challenging patches or patches including areas of particular interests within a training dataset are selected with higher probabilities than others. In some other embodiments, generative adversarial networks (GANs) may be used for training a machine-learning based denoiser as an alternative to using pre-defined loss functions.

公开(公告)号：US20180293710A1

公开(公告)日：2018-10-11

申请号：US15630478

申请日：2017-06-22

申请人： PIXAR

发明人： Mark Meyer ,  Anthony DeRose ,  Steve Bako

IPC分类号： G06T5/00

CPC分类号： G06T5/002 ,  G06T2207/20081 ,  G06T2207/20084

摘要： The present disclosure relates to using a neural network to efficiently denoise images that were generated by a ray tracer. The neural network can be trained using noisy images generated with noisy samples and corresponding denoised or high-sampled images (e.g., many random samples). An input feature to the neural network can include color from pixels of an image. Other input features to the neural network, which would not be known in normal image processing, can include shading normal, depth, albedo, and other characteristics available from a computer-generated scene. After the neural network is trained, a noisy image that the neural network has not seen before can have noise removed without needing manual intervention.