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: Provided herein are methods, systems, and computer products for evaluating nodes concurrently using a modified data flow graph. The modified data flow graph can identify independent nodes that can run as separate tasks. However, rather than relying on declared dependencies, embodiments herein can determine dependencies between segments of data elements in a data flow graph, and modify the data flow graph to take advantage of the determined dependencies. In such embodiments, the data elements can be divided into segments. By separating data elements into segments, nodes that previously depended on each other can be evaluated concurrently when independent segments are identified.

Abstract: 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.

Abstract: Techniques are proposed for embedding transition points in media content. A transition point system retrieves a time marker associated with a point of interest in the media content. The transition point system identifies a first position within the media content corresponding to the point of interest. The transition point system embeds data associated with the time marker into the media content at a second position that is no later in time than the first position. The transition point system causes a client media player to transition from a first image quality level to a second quality level based on the time marker.

Abstract: Provided are methods, systems, and computer-program products for recovering from intersections during a simulation of an animated scene when a collision detection operation is active. For example, the collision detection operation can be selectively activated and deactivated during the simulation of one or more objects for a time step based on an intersection analysis, which can identify intersections of the one or more objects for the time step. Once the collision detection operation is deactivated, a collision response can apply one or more forces to intersecting portions of the one or more objects to eliminate the intersections of the one or more objects. For example, a portion of a cloth that is in a state of intersection can be configured such that the collision detection operation is not performed on the portion, thereby allowing the cloth to be removed from inside of another object by a collision response algorithm.

Abstract: This disclosure provides an approach for automatically generating UV maps for modified three-dimensional (3D) virtual geometry. In one embodiment, a UV generating application may receive original 3D geometry and associated UV panels, as well as modified 3D geometry created by deforming the original 3D geometry. The UV generating application then extracts principal stretches of a mapping between the original 3D geometry and the associated UV panels and transfers the principal stretches, or a function thereof, to a new UV mapping for the modified 3D geometry. Transferring the principal stretches or the function thereof may include iteratively performing the following steps: determining new UV points assuming a fixed affine transformation, determining principal stretches of a transformation between the modified 3D geometry and the determined UV points, and determining a correction of a transformation matrix for each triangle to make the matrix a root of a scoring function.

Abstract: Systems and methods can provide computer animation of animated scenes or interactive graphics sessions. A grid camera separate from the render camera can be created for segments where the configurations (actual or predicted) of the render camera satisfy certain properties, e.g., an amount of change is within a threshold. If a segment is eligible for the use of the separate grid camera, configurations of the grid camera during a segment can be determined, e.g., from the configurations of the render camera. The configurations of the grid camera can then be used to determine grids for rendering objects. If a segment is not eligible for the use of the grid camera, then the configurations of the render camera can be used to determine the grids for rendering.

Abstract: A user interface provides for multi-stroke marking menus and other uses, for use on multitouch devices. One variant of multi-stroke marking is where users draw strokes with either both hands simultaneously or alternating between the hands. Alternating strokes between hands doubles the number of accessible menu items for the same number of strokes. Other inputs can be used as well, such as timing, placement, and direction.

Abstract: Particular embodiments comprise providing a surface mesh for an object, generating a voxel grid comprising volumetric masks for the mesh, and generating a lit mesh, wherein the lit mesh comprises a shaded version of the mesh as positioned in a scene. The voxel grid may be positioned over the lit mesh in the scene, and a first ray may be traced to a position of the voxel grid. If the traced ray passed through the voxel grid and hit a location on the lit mesh, then one or more second rays may be traced to the hit location on the lit mesh. If the traced ray hit a location in the voxel grid but did not hit a location on the lit mesh, then one or more second rays may be traced from the hit location in the voxel grid to the closest locations on the lit mesh. Finally, color sampled at one or more locations proximate to the position of the voxel grid may be blurred outward through the voxel grid to create a volumetric projection.

Abstract: Techniques are disclosed for accounting for features of computer-generated dynamic or simulation models being at different scales. Some examples of dynamic or simulation models may include models representing hair, fur, strings, vines, tails, or the like. In various embodiments, features at different scales in a complex dynamic or simulation model can be treated differently when rendered and/or simulated.

Abstract: Techniques for animating a non-rigid object in a computer graphics environment. A three-dimensional (3D) curve rigging element representing the non-rigid object is defined, the 3D curve rigging element comprising a plurality of knot primitives. One or more defined values are received for an animation control attribute of a first knot primitive. One or more values are generated, for a second animation control attribute for a second knot primitive, based on the plurality of animation control attributes of a neighboring knot primitive. An animation is then rendered using the 3D curve rigging element. More specifically, one or more defined values for the first attribute of the first knot primitive and the generated value for the second attributes of the second knot primitive are used to generate the animation. The rendered animation is output for display.

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: 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: 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: A method and system for rendering a three-dimensional (3D) scene by excluding non-contributing objects are disclosed. A preliminary object analysis using relatively few rays can be performed to determine which off-camera objects are to be excluded or included in the rendering process. The preliminary object analysis may involve performing an initial ray path tracing to identify intersections between a plurality of rays and one or more objects in the 3D scene. The object analysis can include identifying whether a first object in the 3D scene can be identified as an off-camera object. When the first object is identified as an off-camera object, a number of intersections between the plurality of rays and the first object can be counted. If the number of intersections is less than a corresponding threshold, the first object can be identified as being excluded from a future rendering process to render the first frame.

Abstract: 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.

Abstract: Systems, method, and computer program products for compressing a deep image comprising a plurality of voxels by, for each of the plurality of voxels, converting a voxel value to a corresponding value in a Lie algebra based on a logarithmic mapping function, interpolating a first subset of the plurality of values in the Lie algebra using a linear interpolation function applied to a first endpoint and a second endpoint of a first voxel column of the deep image, and upon determining that a deviation of the interpolation of each value in the first subset of the plurality of values does not exceed a threshold, storing an indication of the first endpoint, the second endpoint, and the respective values in the Lie algebra corresponding to the first and second endpoints.

Abstract: Techniques are disclosed for solving geometry processing tasks on a subdivision surface of an input geometry using a subdivision exterior calculus (SEC) framework. A control polygonal mesh is received for generating a subdivision surface model. The polygonal mesh is associated with subdivision levels. To generate the subdivision surface model, one or more subdivision matrices of the polygonal mesh is determined at each subdivision level. One or more SEC matrices is computed from the subdivision matrices. The differential equation required by the geometry processing application is then solved numerically on the input control mesh using the SEC matrices.

Abstract: In various embodiments, a user can create or generate objects to be modeled, simulated, and/or rendered. The user can apply a mesh to the character's form to create the character's topology. Information, such as character rigging, shader and paint data, hairstyles, or the like can be attached to or otherwise associated with the character's topology. A standard or uniform topology can then be generated that allows information associated with the character to be transfer to other characters that have a similar topological correspondence.