Nvidia Corporation

Nikolaus Binder

Alexander Georg Keller

Carsten Alexander Waechter

Ken Dahm

Fabrice Pierre Armand Rousselle

Jan Novák

Thomas Müller

Samuli Matias Laine

Timo Oskari Aila

Tero Tapani Karras

Jaakko Tapani Lehtinen

Lutz Kettner

Marc Droske

Pascal Gautron

Rajko Yasui-Schoeffel

Sascha Fricke

Enzo Catalano

In photorealistic image synthesis by light transport simulation, the colors of each pixel are an integral of a high-dimensional function. However, the functions to integrate contain discontinuities that cannot be predicted efficiently. In practice, the pixel colors are estimated by using Monte Carlo and quasi-Monte Carlo methods to sample light transport paths that connect light sources and cameras and summing up the contributions to evaluate an integral. Because of the sampling, images appear noisy when the number of samples is insufficient. A low discrepancy sequence provides sample locations and these sample locations can be enumerated (assigned or distributed to pixels) according to a space-filling curve superimposed on a pixel grid. Correlations of such combinations of space-filling curves and low discrepancy sequences are analyzed, and the presented algorithms reduce correlations, are deterministic, and may be executed for each pixel in parallel.

Rendering along a space-filling curve

A fully-connected neural network may be configured for execution by a processor as a fully-fused neural network by limiting slow global memory accesses to reading and writing inputs to and outputs from the fully-connected neural network. The computational cost of fully-connected neural networks scale quadratically with its width, whereas its memory traffic scales linearly. Modern graphics processing units typically have much greater computational throughput compared with memory bandwidth, so that for narrow, fully-connected neural networks, the linear memory traffic is the bottleneck. The key to improving performance of the fully-connected neural network is to minimize traffic to slow 'global' memory (off-chip memory and high-level caches) and to fully utilize fast on-chip memory (low-level caches, “shared” memory, and registers), which is achieved by the fully-fused approach. A real-time neural radiance caching technique for path-traced global illumination is implemented using the fully-fused neural network for caching scattered radiance components of global illumination.

Fully-fused neural network execution

A fully-connected neural network may be configured for execution by a processor as a fully-fused neural network by limiting slow global memory accesses to reading and writing inputs to and outputs from the fully-connected neural network. The computational cost of fully-connected neural networks scale quadratically with its width, whereas its memory traffic scales linearly. Modern graphics processing units typically have much greater computational throughput compared with memory bandwidth, so that for narrow, fully-connected neural networks, the linear memory traffic is the bottleneck. The key to improving performance of the fully-connected neural network is to minimize traffic to slow 'global' memory (off-chip memory and high-level caches) and to fully utilize fast on-chip memory (low-level caches, 'shared' memory, and registers), which is achieved by the fully-fused approach. A real-time neural radiance caching technique for path-traced global illumination is implemented using the fully-fused neural network for caching scattered radiance components of global illumination.

An apparatus, computer readable medium, and method are disclosed for decompressing compressed geometric data stored in a lossless compression format. The compressed geometric data resides within a compression block sized according to a system cache line. An indirection technique maps a global identifier value in a linear identifier space to corresponding variable rate compressed data. The apparatus may include decompression circuitry within a graphics processing unit configured to perform ray-tracing.

Block-based lossless compression of geometric data

A method, computer readable medium, and system are disclosed for performing tree traversal with backtracking in constant time. The method includes the steps of traversing a tree, maintaining a bit trail variable and a current key variable during the traversing, where the bit trail variable includes a first plurality of bits indicating tree levels on which a node has been postponed along a path from the root of the tree during the traversing, and the current key variable includes a second plurality of bits indicating a number of a current node within the tree, and performing backtracking within the tree during the traversing, utilizing the bit trail variable and the current key variable.

Tree traversal with backtracking in constant time

A method, computer readable medium, and system are disclosed for reducing noise during a rendering of a scene by sharing information that is spatially close through path space filtering. A vertex of a light transport path is selected, and one or more features of the selected vertex are quantized. A first hash is calculated based on the one or more quantized features of the selected vertex, and a collision resolution is performed within a hash table. A contribution of the light transport path at the selected vertex is accumulated to the hash table, and a counter is incremented in response to adding the contribution of the light transport path at the selected vertex to the hash table. An average contribution of the light transport path is then calculated, utilizing the counter.

Reducing noise during rendering by performing parallel path space filtering utilizing hashing

A method, a computer program, and a production renderer for accelerating a rendering process of an image are provided. In one embodiment, the method includes intercepting a first invocation of a function from a custom shader during a rendering process of an image, computing a result of the function employing a processor, and returning the result to the custom shader in response to a second invocation of the function during the rendering process.

Method and a production renderer for accelerating image rendering

A method and renderer for a progressive computation of a light transport simulation are provided. The method includes the steps of employing a low discrepancy sequence of samples; and scrambling an index of the low discrepancy sequence independently per region using a hash value based on coordinates of a respective region, wherein for each set of a power-of-two number of the samples, the scrambling is a permutation.

Decorrelation of low discrepancy sequences for progressive rendering

A system, method, and computer program product are provided for performing path space filtering. In use, a set of light transport paths associated with a scene is sampled. Additionally, a plurality of vertices associated with the sampled set of light transport paths is selected, where each selected vertex has an associated throughput and light contribution. Further, an averaged light contribution of each of the selected plurality of vertices is determined, utilizing one or more weights. Further still, the averaged light contribution of each of the selected plurality of vertices is combined after multiplying the averaged light contribution of each of the selected vertices by the associated throughput of the vertex.

System, method, and computer program product for performing path space filtering

A system, method, and computer program product are provided for determining a quantity of light received by an element of a scene. In use, a quantity of light received by a first element of the scene is determined by averaging a quantity of light received by elements of the scene that are associated with a selected set of light paths.

System, method, and computer program product for determining a quantity of light received by an element of a scene

A system, method, and computer program product are provided for simulating light transport. In operation, a distribution function is decomposed utilizing a technique for sampling from a probability distribution (e.g. the Alias Method, etc.). Additionally, light transport associated with at least one scene is simulated utilizing information associated with the decomposed distribution function.

System, method, and computer program product for simulating light transport

A system, method, and computer program product are provided for generating a subset of a low discrepancy sequence. In use, a low discrepancy sequence is identified. Additionally, a threshold value is determined. Further, a single dimension of the low discrepancy sequence is selected. Further still, for each element included within the low discrepancy sequence, the selected single dimension is compared to the determined threshold value. Also, a subset of the low discrepancy sequence is generated, based on the comparing.

System, method, and computer program product for generating a subset of a low discrepancy sequence

A system, method, and computer program product are provided for performing ray tracing. In use, ray tracing is performed utilizing a divide and conquer method, where the divide and conquer method is associated with a cache.

System, method, and computer program product for performing ray tracing

A system, method, and computer program product are provided for evaluating an integral utilizing a low discrepancy sequence and a block size. In use, a low discrepancy sequence and a block size are determined. Additionally, an integral is evaluated, utilizing the low discrepancy sequence and the block size.

Growing community of inventors

Berlin, Germany

Nikolaus Binder