Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: Systems, methods, and articles of manufacture for physically-based sculpting of virtual elastic materials are provided. The physically-based sculpting in one embodiment simulates elastic responses to localized distributions of force produced by sculpting with a brush-like force (e.g., grab, twist, pinch, scale) using one or more regularized solutions to equations of linear elasticity applied to a virtual infinite elastic space, referred to herein as “regularized Kelvinlets.

Abstract: Embodiments herein describe a headset that simulates accelerations that correspond to a visual presentation being viewed by the user. The headset includes a force system that applies a force on the head of the user to simulate an acceleration being viewed by the user. The force system may include an actuator that moves a weight to different locations on or around the headset. By moving the weight to different locations, the weight can apply a force that simulates acceleration. For example, the headset can move the weight to apply a force that lifts the head of the user up, which is similar to a force that would be applied if the user was physically accelerated forward. By moving the weight, the force system can simulate accelerations in any number of directions—e.g., front, back, left, right, etc.

Abstract: One embodiment of the present application includes an approach by which an animation system manipulates an animatable object. The animation system detects that a pointer device has positioned a pointer location at a first location, the first location coinciding with a first portion of geometry of the animatable object. The animation system indicates that a first manipulator associated with the first portion of geometry is tentatively selected. Prior to receiving a selection event from the pointer device, the animation system displays a representation of the first manipulator.

Abstract: 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.

Abstract: Techniques for smoothing curves used in computer animation are disclosed. In one embodiment, a smoothing application determines a number of tangents to a curve in response to a modification to a knot or the addition of a new knot, by first determining phantom tangents at knots that are neighbors of each knot that is processed. The smoothing application then (1) determines a length of each side of the tangent at each knot being processed as 1/N times the x-axis distance to a neighboring knot on the same side, (2) determines initial angles of the tangent at each knot being processed by pointing a tip of each side of the tangent at a near tip of a previously determined phantom tangent on the same side, and (3) reconciles the initial angles determined for the tangent at each knot being processed by taking a weighted sum of those initial angles.

Abstract: One aspect of the present disclosure is directed to enabling a user to specify one or more forces to influence how a movable object carried by a 3D character may move during an animation sequence of the 3D character. In some embodiments, the user input can include an arrow. The user can be enabled manipulate the arrow to specify values for at least one parameter of the force to be applied to the movable object during the animation sequence. Another aspect of the disclosure is directed to enabling the user to draw a silhouette stroke to direct an animation of the movable object during the animation sequence. The silhouette stroke drawn by the user can be used as a “boundary” towards which the movable object may be “pulled” during the animation sequence. This may involve generating forces according to the position where the silhouette stroke is drawn.

Abstract: Embodiments can generate content (e.g., a feature film, virtual reality experience) in a virtual environment (e.g., a VR environment). Specifically, they allow fast prototyping and development of a virtual reality experience by allowing virtual assets to be quickly imported into a virtual environment. The virtual assets can be used to help visualize or “storyboard” an item of content during early stages of development. In doing so, the content can be rapidly iterated upon without requiring use of more substantial assets, which can be time consuming and resource intensive.

Abstract: The present disclosure relates to a method, computer program product, and system for rendering one or more objects in a set. The illumination agent designates a region of interest in the set to be rendered. The illumination agent defines an amount of photons to be directed towards the region of interest in the set. The illumination agent generates a photon map of the set. The illumination agent generates a portion of the photon map based on the region of interest and the designated amount of photons to be applied to the region of interest. The illumination agent generates a remainder of the photon map based on an area exterior to the region of interest and a second amount of photons to be applied to the area. The processor transmits the photon map for processing.

Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: Techniques are disclosed for generating quality renderings of volumes by sampling a volume light by generating and analyzing a sparse voxel octree. In one embodiment, a volumetric light source may be divided into voxels and importance information stored in an octree. An importance value may be determined for each voxel based on the amount of emitted light in the region associated with that voxel. Importance values regarding the individual voxels may be stored in the leaves of the octree. Each interior node may be associated with an importance value equal to the sum of the importance values of its children. The root node may be associated with the total importance of the entire octree.

Abstract: User interface display layouts are provided that draw a user's attention to a specific element or elements by de-emphasizing the surrounding content, but without removing the de-emphasized content from the interface. This ability to maintain the whole presentable layout with visibility layers and without layout changes provides a useful navigation experience for the user as it is clear where the user's attention should go and yet the surrounding content is still subtly there, constantly reminding the user of the other available content. De-emphasis of certain content items is achieved by modifying display characteristics of those content items relative to a base display level, for example by lowering saturation, lowering opacity, and/or de-focusing (as if the user is looking through a camera) and modification can be done variably. Driven by a relevancy score, each content item in a display layout can be de-emphasized more or less depending on which content is more meaningful to the user's filtering actions.

Abstract: 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.