Christiaan Gribble
Principal Research Scientist

Applied Technology Operation
SURVICE Engineering Company

Mailing address:
     6101 Penn Avenue, Suite 301
     Pittsburgh, PA  15206

Phone numbers:
     412.342.8219 (office)
     410.272.6763 (fax)

E-mail address:
    

 


[ home ]   [ research ]   [ vita ]

 

2017


     Real-Time, In Situ Intelligent Video Analytics: Harnessing the Power of GPUs for Deep Learning Applications

Shawn Recker and Christiaan Gribble

DSIAC Journal, vol. 4, no. 1

Modern graphics processing units (GPUs), now explicitly designed for deep learning algorithms, have enabled many of the recent and impressive achievements of contemporary deep learning techniques. At the same time, these massively parallel architectures are driving the next generation of deep learning algorithms for intelligent video analytics (IVA). We discuss contemporary approaches to deep learning and highlight Sentinel, a system leveraging modern GPU architectures and advances in deep learning to provide real-time in situ IVA for applications in defense, homeland security, emergency response, and disaster relief, among others.

 
 
2016


     Implementing Node Culling Multi-Hit BVH Traversal in Embree

Christiaan Gribble, Ingo Wald, and Jefferson Amstutz

Journal of Computer Graphics Techniques, vol. 5, no. 4

We present an implementation of node culling multi-hit BVH traversal in Embree. Whereas previous work was limited by API restrictions, as of version 2.10.0, Embree respects ray state modifications from within its intersection callbacks (ICBs). This behavior permits an ICB-based implementation of the node culling algorithm, but requires a clever trick to induce culling once the necessary conditions are met: candidate intersections are accepted with a (possibly) modified tfar value, so that ray traversal continues with an updated ray interval. We highlight a scalar implementation of the Embree ICB and report results for a vector implementation in OSPRay that improves performance by as much as 2x relative to naive multi-hit when users request fewer-than-all hits.
     
     Realizing Multi-Hit Ray Tracing in Embree & OSPRay

Christiaan Gribble

Technical session, Intel HPC Developer Conference, November 2016

We discuss multi-hit ray traversal, a class of ray traversal algorithms that finds one or more, and possibly all, primitives intersected by a ray and ordered by point of intersection. After a brief review the multi-hit problem, we discuss the implementation of high performance multi-hit ray traversal in Embree---Intel's high performance ray tracing kernels---and in OSPRay---Intel's ray based rendering engine for high fidelity visualization---using the Intel SPMD Program Compiler, ispc. We then demonstrate our implementation with two key applications:
  1. fast and accurate rendering of transparent objects for high fidelity visualization, and

  2. a physics-based simulation in which the interesting phenomena are governed by equations similar to the Beer-Lambert Law.
In the first, our multi-hit ray tracing module for OSPRay is used directly for visualization, demonstrating the utility of OSPRay for high performance, high fidelity visualization of complex engineering CAD models. In the second, our module is used both for simulation and for visualization, further demonstrating OSPRay's flexibility. Both applications showcase OSPRay as an extensible framework in which to implement simulation and visualization techniques supporting scalable, high performance visual analysis tools across several domains.
     
     Real-Time In-Situ Intelligent Video Analytics for Mobile Platforms

Christiaan Gribble

Technical session, GPU Technology Conference Washington, D.C., October 2016

We highlight Sentinel, a system for real-time in-situ intelligent video analytics on mobile computing platforms. Sentinel combines state-of-the-art techniques in HPC with Dynamic Mode Decomposition (DMD), a proven method for data reduction and analysis. By leveraging CUDA, our early system prototype achieves significantly better-than-real-time performance for DMD-based background/foreground separation on high-definition video streams, thereby establishing the efficacy of DMD as the foundation on which to build higher level real-time computer vision techniques. In this talk, we present an overview of the Sentinel system, including the application of DMD to background/foreground separation in video streams, and outline our current efforts to enhance and extend the prototype system.
     
     HawkEye: Automatic Stitching of Hand-Held LIDAR Scans using Photogrammetry and Structure-from-Motion

Shawn Recker, Rob Baltrusch, Christiaan Gribble, and Mark Butkiewicz

Coordinate Metrology Systems Conference, July 2016

The reconstruction of a scene from multiple images or video streams has become an essential component in many modern applications. This paper presents a hand-held consumer-grade optical scanner and supporting application suite, called HawkEye. The fully automatic reconstruction system is based upon advancements in hybrid photogrammetry/structure-from-motion technologies and produces a globally aligned 3D model. Analysis of HawkEye results demonstrate visually accurate scene reconstructions.
     
     Node Culling Multi-Hit BVH Traversal

Christiaan Gribble

Eurographics Symposium on Rendering (EI&I track), June 2016

We introduce node culling multi-hit BVH traversal to enable faster multi-hit ray tracing in a bounding volume hierarchy (BVH). Existing, widely used ray tracing engines expose API features that enable implementation of multi-hit traversal without modifying their underlying---and highly optimized---BVH construction and traversal routines; however, this approach requires naive multi-hit traversal to guarantee correctness. We evaluate two low-overhead, minimally invasive, and flexible API mechanisms that enable node culling implementation entirely with user-level code, thereby leveraging existing BVH construction and traversal routines. Results show that node culling offers potentially significant improvement in multi-hit performance in a BVH for cases in which users request fewer-than-all hits.

Supplemental materials, including source code for the reference implementation, are also available.

Unless otherwise stated directly in the source, the code is distributed under the BSD 3-Clause License (see LICENSE for more information). Instructions for building and running the mhBVH driver program are available in the README.

 
 
2015


     An Evaluation of Multi-Hit Ray Traversal in a BVH using Existing First-Hit/Any-Hit Kernels

Jefferson Amstutz, Christiaan Gribble, Johannes Gunther, and Ingo Wald

Journal of Computer Graphics Techniques, vol. 4, no. 4

We explore techniques for multi-hit ray tracing in a bounding volume hierarchy (BVH) using existing ray traversal kernels and intersection callbacks. BVHs are problematic for implementing multi-hit ray traversal correctly due to the potential for spatially overlapping leaf nodes. We demonstrate that the intersection callback feature of modern, high performance ray-tracing APIs enable correct and efficient implementation of multi-hit ray tracing despite this concern. Moreover, the callback-based approach enables multi-hit ray tracing using existing, highly optimized BVH data structures, mitigating maintenance issues imposed by hand-tuned multi-hit traversal kernels across various hardware architectures. Results show that memory-bandwidth limitations and SIMD vector width of the target hardware platform dictate ideal hit-point memory layout, as well as the point at which sorting should occur, in order to maximize performance with existing BVH traversal algorithms.

Slides for the paper presentation at ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games 2016 are also available.
     
     Effective Parallelization Strategies for Scalable, High Performance Radio Frequency Ray Tracing

Christiaan Gribble and Jefferson Amstutz

IEEE High Performance Extreme Computing, September 2015

We present StingRay, an interactive environment for combined RF simulation and visualization based on ray tracing. StingRay is explicitly designed to support scalable, high performance simulation and visualization of RF energy propagation in complex urban environments using modern, highly parallel computer architectures. We explore three strategies for exploiting parallelism in StingRay and provide evaluations of their scalability and performance on a modern workstation-class system. Results show that a more scalable, higher performing version of StingRay is possible with careful attention to the expression of task-level parallelism in OpenMP.
     
     Radio Frequency Ray Tracing

Christiaan Gribble and Jefferson Amstutz

Book chapter, High Performance Parallelism Pearls, Volume 2, July 2015

Radio frequency simulation and visualization is critical to planning, analyzing, and optimizing wireless communication networks. An interactive tool supporting visual analysis of RF propagation characteristics in complex environments will enable analysts to better understand RF propagation phenomena. We discuss StingRay, an interactive combined RF simulation and visualization environment that satisfies these constraints. Derived from first principles, StingRay's ray-based RF simulation engine provides a high fidelity approach for RF prediction in complex environments. StingRay is explicitly designed for Intel Xeon processors and Intel Xeon Phi coprocessors, which provide high performance compute capabilities for physics based simulation. In this way, StingRay combines best-known methods in high performance ray tracing and high fidelity visualization with low-level, architecture-specific optimizations for Intel Xeon processors and Intel Xeon Phi coprocessors to provide an interactive environment for predictive simulation and visualization of RF energy propagation in complex environments.
     
     An Evaluation of Existing BVH Traversal Algorithms for Efficient Multi-Hit Ray Tracing

Jefferson Amstutz, Johannes Guenther, Ingo Wald, and Christiaan Gribble

Poster, ACM SIGGRAPH/Eurographics High Performance Graphics, July 2015

We explore techniques for multi-hit ray tracing in a bounding volume hierarchy (BVH). BVHs are problematic for implementing multi-hit ray traversal correctly due to the potential for spatially overlapping leaf nodes. We demonstrate that the intersection callback feature of modern, high performance ray tracing APIs enables correct and efficient implementation of multi-hit ray tracing despite this concern. Moreover, the callback-based approach enables multi-hit ray tracing using existing, highly optimized BVH data structures, mitigating maintenance issues imposed by hand-tuned multi-hit traversal kernels across various hardware architectures. Results show that memory bandwidth limitations and SIMD vector width of the target hardware platform dictate ideal hit point memory layout, as well as the point at which sorting should occur, to maximize performance with existing BVH traversal algorithms.

An extended abstract for the poster and slides for the corresponding quick talk are also available.
     
     StingRay: High-Performance RF Energy Propagation Modeling in Complex Environments

Christiaan Gribble and Jefferson Amstutz

DSIAC Journal, vol. 2, no. 2

StingRay is an interactive environment for combined radio frequency (RF) simulation and visualization using ray tracing. RF ray tracing offers an alternative to traditional RF simulation methods, and provides several advantages over these methods. StingRay leverages these advantages to provide highly interactive visual analysis of RF propagation characteristics. StingRay achieves high performance RF simulation by leveraging Intel's Embree ray tracing kernels, and Intel's OSPRay rendering engine provides high fidelity visualization of the resulting data. We show that StingRay enables high performance, high fidelity combined simulation and visualization of RF energy propagation, providing the ability to quickly identify phenomena of interest, thereby reducing time-to-insight for many mission planning tasks.
     
     Leveraging GPUs for Ballistic Simulation

Jefferson Amstutz and Christiaan Gribble

DSIAC Journal, vol. 2, no. 1

Massively parallel coprocessors, specifically graphics processing units (GPUs), are becoming increasingly popular as a means to accelerate computation. However, the ballistic simulation community has seen relatively little adoption of GPUs into their workflows, due in part to the complexity of mapping existing codes to these devices. We discuss a new software architecture for ballistic simulation that effectively exploits GPUs, resulting in the ability to execute simulations in real-time on a single workstation.

 
 
2014


     High Performance Radio Frequency Ray Tracing with Embree

Christiaan Gribble and Jefferson Amstutz

Technical session, Intel HPC Developer Conference, November 2014

Predictive simulation and visualization of radio frequency (RF) energy propagation is critical to planning, analyzing, and optimizing wireless communication networks. RF ray tracing (RFRT) is an alternative to traditional RF simulation methods that offers several advantages over these methods. First, implementation of ray-based RF energy transport is fairly straightforward. Second, RFRT generates full signal trajectories, allowing computation and visualization of signal characteristics that are extremely costly, or even impossible, with other methods. Finally, RFRT scales effectively, both with geometric complexity and with core count.

This talk presents the details of StingRay, an interactive environment for combined RF simulation and visualization. High performance RF simulation is achieved by leveraging Intel's Embree ray tracing kernels, and Intel's OSPRay rendering engine provides high fidelity visualization of the resulting data. We show that, together, StingRay, Embree, and OSPRay enable high performance, high fidelity combined simulation and visualization of physical phenomena in non-optical domains, including RF energy propagation.

An abbreviated version of this talk is also featured in the Intel Parallel Universe Theater, part of the Supercomputing 2014 Exhibition.
     
     Depth Data Assisted Structure-from-Motion Parameter Optimization and Feature Track Correction

Shawn Recker, Christiaan Gribble, Mikhail M. Shashkov, Mario Yepez, Mauricio Hess-Flores, and Kenneth I. Joy

Applied Imagery Pattern Recognition Workshop, October 2014

Structure-from-Motion (SfM) applications attempt to reconstruct the three-dimensional (3D) geometry of an underlying scene from a collection of images, taken from various camera viewpoints. Traditional optimization techniques in SfM, which compute and refine camera poses and 3D structure, rely only on feature tracks, or sets of corresponding pixels, generated from color (RGB) images. With the abundance of reliable depth sensor information, these optimization procedures can be augmented to increase the accuracy of reconstruction. This paper presents a general cost function, which evaluates the quality of a reconstruction based upon a previously established angular cost function and depth data estimates. The cost function takes into account two error measures: first, the angular error between each computed 3D scene point and its corresponding feature track location, and second, the difference between the sensor depth value and its computed estimate. A bundle adjustment parameter optimization is implemented using the proposed cost function and evaluated for accuracy and performance. As opposed to traditional bundle adjustment, in the event of feature tracking errors, a corrective routine is also present to detect and correct inaccurate feature tracks. The filtering algorithm involves clustering depth estimates of the same scene point and observing the difference between the depth point estimates and the triangulated 3D point. Results on both real and synthetic data are presented and show that reconstruction accuracy is improved.
     
     Hybrid Photogrammetry Structure-from-Motion Systems for Scene Measurement and Analysis

Shawn Recker, Mikhail M. Shashkov, Mauricio Hess-Flores, Christiaan Gribble, Rob Baltrusch, Mark A. Butkiewicz, and Kenneth I. Joy

Coordinate Metrology Systems Conference, July 2014

Given the recent advances in both photogrammetry and structure-from-motion, a pipeline that capitalizes on the strengths of both fields is now possible. This paper presents a hybrid system that uses photogrammetric information to improve the accuracy of structure-from-motion, which in turn provides a more dense reconstruction. The procedure maintains the required metrological accuracy and permits measurements between points with no corresponding targets. The paper provides an analysis of the effects of various camera parameters to determine optimal scene configurations. Results generated by the hybrid system for real and synthetic data demonstrate that both more accurate and more dense reconstructions are obtained than with structure-from-motion alone.

This paper also appears as a feature article in CMSC World, 13 November 2014.
     
     Multi-Hit Ray Traversal

Christiaan Gribble, Alexis Naveros, and Ethan Kerzner

Journal of Computer Graphics Techniques, vol. 3, no. 1

Multi-hit ray traversal is a class of ray traversal algorithms that finds one or more, and possibly all, primitives intersected by a ray ordered by point of intersection. Multi-hit traversal generalizes traditional first-hit ray traversal and is useful in computer graphics and physics-based simulation. We introduce an efficient algorithm for ordered multi-hit ray traversal, investigate its performance in a GPU ray tracer, and demonstrate two problems easily solved with our algorithm.

Slides for the paper presentation at ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games 2014 are also available.

 
 
2013


     Ray Tracing is the Future and Ever Will Be...

Alexander Keller, Ingo Wald, Tero Karras, Samuli Laine, Jacco Bikker, Christiaan Gribble, Won-Jong Lee, and James McCombe

Course, ACM SIGGRAPH, July 2013

The primary objective of this course is to present a coherent summary of the state of the art in ray tracing technology. The course covers the most recent developments and practical aspects of the parallel construction of acceleration data structures and traversal of such acceleration data structures using highly parallel processors, including a discussion of divergent code paths and memory accesses as well as occupancy. Ray tracing in real-time games is considered one of the main application opportunities, but an important part of the course focuses on hardware for ray tracing applications in mobile platforms.
     
    GPU Ray Tracing with Rayforce

Christiaan Gribble and Alexis Naveros

Poster, ACM SIGGRAPH, July 2013

Rayforce is a high performance ray tracing engine designed for massively parallel computing architectures, including manycore GPUs. Rayforce leverages a novel graph-based acceleration structure that permits first-hit, any-hit, and multi-hit traversal algorithms required to solve a variety of problems in physics-based simulation domains. Rayforce exposes core ray tracing operations via a programmable interface to enable the implementation of various computer graphics and scientific computing applications.

An extended abstract for the poster is also available.
     
    Toward Accurate and Efficient Order-Independent Transparency

Ethan Kerzner, Chris Wyman, Lee Butler, and Christiaan Gribble

Poster, ACM SIGGRAPH, July 2013

Correctly rendering multi-layered transparent geometry requires accumulating contributions from multiple fragments per pixel. Dynamic A-buffers (e.g., per-pixel linked lists [Yang et al. 2010]) achieve this by storing and sorting fragments on-the-fly. We introduce an improvement to recent GPU-based interactive A-buffer techniques: we decouple visibility and shading to reduce memory demands of multi-fragment rendering.

An extended abstract for the poster is also available.
     
    Advances in High-performance GPU Ray Tracing for Physics-based Simulation

Christiaan Gribble and Lee Butler

Technical session, GPU Technology Conference, March 2013

Explore recent advances in high-performance GPU ray tracing for applications other than optical rendering. In this session, we dive into the details of Rayforce, a CUDA ray tracing engine that leverages a new graph-based spatial indexing structure to achieve performance in excess of one billion rays per second in some non-trivial scenarios. We then explore several example applications that leverage Rayforce in a framework for cognition-driven simulation (CDS) that enables analysts to experience an immersive physics-based simulation environment. Compared to traditional means of analysis, CDS represents a next-generation approach to simulation and analysis across a broad range of application domains.
     
    Visual Simulation Laboratory

Lee Butler, Christiaan Gribble, and Mark Butkiewicz

Poster, GPU Technology Conference, March 2013

The Visual Simulation Laboratory (VSL) is an ongoing, open-source framework developed by the U.S. Army Research Laboratory and its collaborators to bring the power of GPU computing to a variety of DoD application domains. VSL is designed to transform legacy workflows into immersive, end-to-end physics-based simulation and analysis tools.

In the new generation of high-performance simulations, visuals must lead the user toward understanding as the simulation progresses. By exposing the user to the process and progress of the simulation, cognition proceeds in tandem with simulation. In this approach, simulation and analysis overlap, allowing the simulation to be driven by the user. This overlap improves understanding of not only the outcome, but of its origin as well, which together improves confidence in the results. Finally, users can identify critical outcomes more quickly and thereby refine the next stimulation step accordingly. We call this approach cognition-driven simulation (CDS), and it enables a new class of physics-based simulations that require substantial computational horsepower.
     
    GPU Ray Tracing with Rayforce

Alexis Naveros and Christiaan Gribble

Poster, GPU Technology Conference, March 2013

Rayforce is a high performance ray tracing engine designed for highly parallel computing architectures including manycore GPUs, multicore CPUs, and hybrid CPU/GPU processors. Rayforce leverages a novel graph-based acceleration structure that permits high-performance first-hit, any-hit, and multi-hit traversal algorithms required to solve a variety of problems in rendering and physically based simulation. These building blocks are exposed via a programmable interface to enable the implementation of computer graphics and scientific computing applications.
     
    Simulating Radio Frequency Propagation via Ray Tracing

Konstantin Shkurko, Thiago Ize, Christiaan Gribble, Erik Brunvand, and Lee Butler

Poster, GPU Technology Conference, March 2013

Prediction of radio frequency (RF) energy propagation in the presence of complex outdoor terrain features—urban environments, for example—is of great interest when planning, optimizing and analyzing wireless networks. A tool for fast prediction could improve network coverage, provide estimates of signal strength throughout the environment, estimate time delay of multipath signals, and provide data for power allocation in the deployed transmitters. Such a tool is essential when planning networks that need to be set up quickly for temporary purposes.

 

[ 2001-2006 ]   [ 2007-2012 ]   [ 2013-present ]

 

Copyright © 2000-2017 Christiaan Gribble
All rights reserved