Jacopo Pantaleoni

Simpler and faster HLBVH with work queues

A recently developed algorithm called Hierachical Linear Bounding Volume Hierarchies (HLBVH) has demonstrated the feasibility of reconstructing the spatial index needed for ray tracing in real-time, even in the presence of millions of fully dynamic triangles. In this work we present a simpler and faster variant of HLBVH, where all the complex book-keeping of prefix sums, compaction and partial breadth-first tree traversal needed for spatial partitioning has been replaced with an elegant pipeline built on top of efficient work queues and binary search. The new algorithm is both faster and more memory efficient, removing the need for temporary storage of geometry data for intermediate computations. Finally, the same pipeline has been extended to parallelize the construction of the top-level SAH optimized tree on the GPU, eliminating round-trips to the CPU, accelerating the overall construction speed by a factor of 5 to 10x.

A path space extension for robust light transport simulation

We present a new sampling space for light transport paths that makes it possible to describe Monte Carlo path integration and photon density estimation in the same framework. A key contribution of our paper is the introduction of vertex perturbations, which extends the space of paths with loosely coupled connections. The new framework enables the computation of path probabilities in the same space under the same measure, which allows us to use multiple importance sampling to combine Monte Carlo path integration and photon density estimation. The resulting algorithm, unified path sampling, can robustly render complex combinations and glossy surfaces and caustics that are problematic for existing light transport simulation methods.

HLBVH: hierarchical LBVH construction for real-time ray tracing of dynamic geometry

We present HLBVH and SAH-optimized HLBVH, two high performance BVH construction algorithms targeting real-time ray tracing of dynamic geometry. HLBVH provides a novel hierarchical formulation of the LBVH algorithm [LGS*09] and SAH-optimized HLBVH uses a new combination of HLBVH and the greedy surface area heuristic algorithm. These algorithms minimize work and memory bandwidth usage by extracting and exploiting coarse-grained spatial coherence already available in the input meshes. As such, they are well-suited for sorting dynamic geometry, in which the mesh to be sorted at a given time step can be defined as a transformation of a mesh that has been already sorted at the previous time step. Our algorithms always perform full resorting, unlike previous approaches based on refitting. As a result they remain efficient even during chaotic and discontinuous transformations, such as fracture or explosion.

VoxelPipe: a programmable pipeline for 3D voxelization

We present a highly flexible and efficient software pipeline for programmable triangle voxelization. The pipeline, entirely written in CUDA, supports both fully conservative and thin voxelizations, multiple boolean, floating point, vector-typed render targets, user-defined vertex and fragment shaders, and a bucketing mode which can be used to generate 3D A-buffers containing the entire list of fragments belonging to each voxel. For maximum efficiency, voxelization is implemented as a sort-middle tile-based rasterizer, while the A-buffer mode, essentially performing 3D binning of triangles over uniform grids, uses a sort-last pipeline. Despite its major flexibility, the performance of our tile-based rasterizer is always competitive with and sometimes more than an order of magnitude superior to that of state-of-the-art binary voxelizers, whereas our bucketing system is up to 4 times faster than previous implementations. In both cases the results have been achieved through the use of careful load-balancing and high performance sorting primitives.

PantaRay: fast ray-traced occlusion caching of massive scenes

We describe the architecture of a novel system for precomputing sparse directional occlusion caches. These caches are used for accelerating a fast cinematic lighting pipeline that works in the spherical harmonics domain. The system was used as a primary lighting technology in the movie Avatar, and is able to efficiently handle massive scenes of unprecedented complexity through the use of a flexible, stream-based geometry processing architecture, a novel out-of-core algorithm for creating efficient ray tracing acceleration structures, and a novel out-of-core GPU ray tracing algorithm for the computation of directional occlusion and spherical integrals at arbitrary points.

Charted metropolis light transport

In this manuscript, inspired by a simpler reformulation of primary sample space Metropolis light transport, we derive a novel family of general Markov chain Monte Carlo algorithms called charted Metropolis-Hastings, that introduces the notion of sampling charts to extend a given sampling domain and make it easier to sample the desired target distribution and escape from local maxima through coordinate changes. We further apply the novel algorithms to light transport simulation, obtaining a new type of algorithm called charted Metropolis light transport, that can be seen as a bridge between primary sample space and path space Metropolis light transport. The new algorithms require to provide only right inverses of the sampling functions, a property that we believe crucial to make them practical in the context of light transport simulation.

Matrix Bidirectional Path Tracing

Sampled paths in Monte Carlo ray tracing can be arbitrarily close to each other due to its stochastic nature. Such clumped samples in the path space tend to contribute little toward an accurate estimate of each pixel. Bidirectional light transport methods make this issue further complicated since connecting paths of sampled subpaths can be arbitrarily clumped again. We propose a matrix formulation of bidirectional light transport that enables stratification (and low-discrepancy sampling) in this connection space. This stratification allows us to distribute computation evenly across contributing paths in the image, which is not possible with standard bidirectional or Markov chain solutions. Each element in our matrix formulation represents a pair of connected eye- and light-subpaths. By carefully reordering these elements, we build a 2D space where equally contributing paths are distributed coherently. We devise an unbiased rendering algorithm that leverages this coherence to effectively sample path space, consistently achieving a 2 − 3 x speedup in radiometrically complex scenes compared to the state-of-the-art.