Patents by Inventor Fabrice Rousselle
Fabrice Rousselle has filed for patents to protect the following inventions. This listing includes patent applications that are pending as well as patents that have already been granted by the United States Patent and Trademark Office (USPTO).
-
Publication number: 20230083929Abstract: 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.Type: ApplicationFiled: November 9, 2022Publication date: March 16, 2023Applicants: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill
-
Patent number: 11532073Abstract: 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.Type: GrantFiled: July 31, 2018Date of Patent: December 20, 2022Assignees: Pixar, Disnev Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill
-
Patent number: 11037274Abstract: 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.Type: GrantFiled: February 12, 2020Date of Patent: June 15, 2021Assignees: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Mark Meyer, Jan Novak
-
Patent number: 10796414Abstract: Supervised machine learning using convolutional neural network (CNN) is applied to denoising images rendered by MC path tracing. The input image data may include pixel color and its variance, as well as a set of auxiliary buffers that encode scene information (e.g., surface normal, albedo, depth, and their corresponding variances). In some embodiments, a CNN directly predicts the final denoised pixel value as a highly non-linear combination of the input features. In some other embodiments, a kernel-prediction neural network uses a CNN to estimate the local weighting kernels, which are used to compute each denoised pixel from its neighbors. In some embodiments, the input image can be decomposed into diffuse and specular components. The diffuse and specular components are then independently preprocessed, filtered, and postprocessed, before recombining them to obtain a final denoised image.Type: GrantFiled: September 26, 2019Date of Patent: October 6, 2020Assignees: Disney Enterprises, Inc., ETH Zürich (Eidgenössische Technische Hochschule Zürich)Inventors: Thijs Vogels, Jan Novák, Fabrice Rousselle, Brian McWilliams
-
Patent number: 10789686Abstract: 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.Type: GrantFiled: January 6, 2020Date of Patent: September 29, 2020Assignees: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Mark Meyer, Jan Novak
-
Patent number: 10706508Abstract: 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.Type: GrantFiled: July 31, 2018Date of Patent: July 7, 2020Assignees: Disney Enterprises, Inc., PixarInventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill
-
Patent number: 10699382Abstract: 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.Type: GrantFiled: July 31, 2018Date of Patent: June 30, 2020Assignees: Disney Enterprises, Inc., PixarInventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill
-
Publication number: 20200184313Abstract: 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.Type: ApplicationFiled: July 31, 2018Publication date: June 11, 2020Applicants: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill
-
Publication number: 20200184605Abstract: 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.Type: ApplicationFiled: February 12, 2020Publication date: June 11, 2020Applicants: PIXAR, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Mark Meyer, Jan Novak
-
Patent number: 10672109Abstract: 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.Type: GrantFiled: July 31, 2018Date of Patent: June 2, 2020Assignees: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill
-
Publication number: 20200143522Abstract: 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.Type: ApplicationFiled: January 6, 2020Publication date: May 7, 2020Applicants: PIXAR, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Mark Meyer, Jan Novak
-
Patent number: 10607319Abstract: 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.Type: GrantFiled: April 5, 2018Date of Patent: March 31, 2020Assignees: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Mark Meyer, Jan Novak
-
Patent number: 10586310Abstract: 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.Type: GrantFiled: April 5, 2018Date of Patent: March 10, 2020Assignees: Pixar, Disney EnterprisesInventors: Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Mark Meyer, Jan Novak
-
Patent number: 10572979Abstract: 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.Type: GrantFiled: April 5, 2018Date of Patent: February 25, 2020Assignees: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Mark Meyer, Jan Novak
-
Publication number: 20200027198Abstract: Supervised machine learning using convolutional neural network (CNN) is applied to denoising images rendered by MC path tracing. The input image data may include pixel color and its variance, as well as a set of auxiliary buffers that encode scene information (e.g., surface normal, albedo, depth, and their corresponding variances). In some embodiments, a CNN directly predicts the final denoised pixel value as a highly non-linear combination of the input features. In some other embodiments, a kernel-prediction neural network uses a CNN to estimate the local weighting kernels, which are used to compute each denoised pixel from its neighbors. In some embodiments, the input image can be decomposed into diffuse and specular components. The diffuse and specular components are then independently preprocessed, filtered, and postprocessed, before recombining them to obtain a final denoised image.Type: ApplicationFiled: September 26, 2019Publication date: January 23, 2020Applicants: Disney Enterprises, Inc., ETH ZÜRICH (EIDGENÖSSISCHE TECHNISCHE HOCHSCHULE ZÜRICH)Inventors: Thijs Vogels, Jan Novák, Fabrice Rousselle, Brian McWilliams
-
Patent number: 10475165Abstract: Supervised machine learning using convolutional neural network (CNN) is applied to denoising images rendered by MC path tracing. The input image data may include pixel color and its variance, as well as a set of auxiliary buffers that encode scene information (e.g., surface normal, albedo, depth, and their corresponding variances). In some embodiments, a CNN directly predicts the final denoised pixel value as a highly non-linear combination of the input features. In some other embodiments, a kernel-prediction neural network uses a CNN to estimate the local weighting kernels, which are used to compute each denoised pixel from its neighbors. In some embodiments, the input image can be decomposed into diffuse and specular components. The diffuse and specular components are then independently preprocessed, filtered, and postprocessed, before recombining them to obtain a final denoised image.Type: GrantFiled: November 15, 2017Date of Patent: November 12, 2019Assignees: Disney Enterprises, Inc., ETH Zürich (Eidgenössische Technische Hochschule ZürichInventors: Thijs Vogels, Jan Novák, Fabrice Rousselle, Brian McWilliams
-
Publication number: 20190304069Abstract: 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.Type: ApplicationFiled: July 31, 2018Publication date: October 3, 2019Applicants: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill, David Adler
-
Publication number: 20190304067Abstract: 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.Type: ApplicationFiled: July 31, 2018Publication date: October 3, 2019Applicants: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill, David Adler
-
Publication number: 20190304068Abstract: 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.Type: ApplicationFiled: July 31, 2018Publication date: October 3, 2019Applicants: Pixar, Disney Enterprises, Inc.Inventors: Thijs Vogels, Fabrice Rousselle, Jan Novak, Brian McWilliams, Mark Meyer, Alex Harvill
-
Publication number: 20180293711Abstract: Supervised machine learning using convolutional neural network (CNN) is applied to denoising images rendered by MC path tracing. The input image data may include pixel color and its variance, as well as a set of auxiliary buffers that encode scene information (e.g., surface normal, albedo, depth, and their corresponding variances). In some embodiments, a CNN directly predicts the final denoised pixel value as a highly non-linear combination of the input features. In some other embodiments, a kernel-prediction neural network uses a CNN to estimate the local weighting kernels, which are used to compute each denoised pixel from its neighbors. In some embodiments, the input image can be decomposed into diffuse and specular components. The diffuse and specular components are then independently preprocessed, filtered, and postprocessed, before recombining them to obtain a final denoised image.Type: ApplicationFiled: November 15, 2017Publication date: October 11, 2018Applicants: Disney Enterprises, Inc., ETH Zürich (Eidgenössische Technische Hochschule Zürich)Inventors: Thijs Vogels, Jan Novák, Fabrice Rousselle, Brian McWilliams