Ray Tracing Visualization Toolkit

Particpants:

  • Christiaan Gribble (faculty)
  • Jeremy Fisher (student, project leader)
  • Daniel Eby (student)
  • Ed Quigley (student)
  • Frank Serra (student, project leader)
  • Andrew Claudy (student)
  • Gideon Ludwig (student)

Description:

We introduce the Ray Tracing Visualization Toolkit (rtVTK), a collection of programming and visualization tools supporting visual analysis of ray-based rendering algorithms. rtVTK comprises a library for recording and processing ray state, together with a flexible software architecture for visualization components, integrated via an extensible GUI. rtVTK enables an investigator to inspect, interrogate, and interact with the computational elements of the ray tracing algorithm itself, thereby promoting a deeper understanding of how computation proceeds. Our goal is to employ real-time ray tracing for applications in fields as diverse as science, engineering, history, and the arts. Many of these applications require predictive images, or those in which computer-generated results are identical to the photo- and radio- metric values obtained by measuring a scene in the physical world. Ray-based rendering algorithms are ideally suited to this task. Typically, these algorithms simulate the behavior of photons as they interact with objects in an environment according to the laws of geometric optics. These interactions are often very complex, and depend on the spatial arrangement of objects in a scene, their material properties, and the optical effects captured by the particular algorithm in use. Moreover, generating a high-fidelity result requires computing many millions (if not billions) of ray/object interactions. Thus, the complexity encountered in predictive rendering applications limits the practicality of current approaches for real- time image synthesis. Even with recent advances targeting highly parallel platforms [van Antwerpen 2011], many seconds of computation are required for results to converge. As such, rendering predictive images at real-time frame rates is not yet feasible. Importantly, predictive applications also require that advanced ray- based rendering algorithms be physically correct for the results to be effective---much more so than applications requiring simply plausible or visually convincing results. Here, too, the complexity of scene geometry, material properties, and optical effects used in predictive rendering leads to difficulties in ensuring program correctness: traditional software debugging tools are not designed to leverage the inherent visual nature of computer graphics computation, and thus lead to a cumbersome debugging process. We believe that effective visualization of ray tracing state will promote deeper understanding of how computation proceeds, addressing a wide variety of problems in ray-based rendering. For example, a subtle and long-standing bug related to secondary ray generation in a batch renderer has been exposed as a result of visualization with rtVTK. Similarly, anecdotal evidence from an undergraduate computer graphics course suggests that students of ray tracing are better able to grasp the algorithm’s details by interacting with a visual representation of the computation. Moreover, visualization with rtVTK may enhance tasks in ray tracing performance analysis by enabling insights beyond the summary statistics provided by traditional analysis tools. Finally, because of its flexibility and extensibility, we believe that rtVTK will be well-suited to new, possibly unforeseen, problem domains and application areas as well. See http://www.rtvtk.org

References:

van Antwerpen , D. 2011. Improving SIMD efficiency for par allel Monte Carlo light transport on the GPU. In Proceedings of High Performance Graphics 2011, 41–50.