Presentation is loading. Please wait.

Presentation is loading. Please wait.

Ray Tracing Dynamic Scenes using Selective Restructuring Sung-eui Yoon Sean Curtis Dinesh Manocha Univ. of North Carolina at Chapel Hill Lawrence Livermore.

Similar presentations


Presentation on theme: "Ray Tracing Dynamic Scenes using Selective Restructuring Sung-eui Yoon Sean Curtis Dinesh Manocha Univ. of North Carolina at Chapel Hill Lawrence Livermore."— Presentation transcript:

1 Ray Tracing Dynamic Scenes using Selective Restructuring Sung-eui Yoon Sean Curtis Dinesh Manocha Univ. of North Carolina at Chapel Hill Lawrence Livermore National Lab

2 Motivations l Dynamic scenes are widely used n Movies, VR applications, and games l Complex and large dynamic scenes n E.g, high-resolution explosion, tears, and fractures

3 An Example of Exploding Dragon (252K triangles)

4 Ray Tracing Dynamic Scenes l Acceleration hierarchy construction n e.g., kd-trees, bounding volume hierarchies, grids, etc l Hierarchy traversal n Perform ray-triangle intersection tests l Key issue n Update the hierarchy as triangles deform

5 Bounding Volume Hierarchies (BVH) based Ray Tracing l Employed early in [Whitted 80] n kd-trees and grids became popular for static models in 90’s l Recently get renewed interest in ray tracing dynamic scenes [Wald et al. 07, Lauterbach et al. 07, Larsson et al. 03] n Simple, but efficient BVH update method is available n Can have better performance

6 BVHs l Object partitioning hierarchies n Uses axis-aligned bounding boxes n Considers surface-area heuristic (SAH) [Goldsmith and Salmon 87] A BVH

7 Two BVH Update Methods Frame 1 Frame 2 BV refitting BV reconstruction O(n log n) O(n log n) Good-quality BVs Good-quality BVs O(n) O(n) Poor-quality BVs Poor-quality BVs

8 Our Goal l Existing BVH update methods n Work at particular classes of dynamic scenes l Design a robust BVH update method n Works well with wide classes of dynamic scenes n Improves the performance of ray tracing

9 Our Contributions l Proposes a novel algorithm to selectively restructure BVHs n Selective restructuring operations n Two probabilistic metrics: culling efficiency and restructuring benefit BVH Restructure Refit

10 Example of Exploding Dragon Model BV refitting Complete reconstruction # of intersections Ray tracing time (sec): construction + traversal Selective restructuring

11 Runtime Captured Video – BART Model (65K triangles) l Compared with the BV refitting method Enabled primary & shadow rays Single thread

12 Prior Hybrid BVH Update Methods l Based on simple heuristics n RT-Deform [Lauterbach et al. 06] n LM method [Larsson and Akenine-Möller 06] l Hard to see what dynamic scenarios they work with or not

13 Runtime Captured Video – BART Model l Compared with RT-Deform Single thread

14 Probabilistic BVH Metrics for Ray Tracing l Culling efficiency n Quantifies the quality of any sub-BVHs n Measures the expected # of intersection tests for a ray l Restructuring benefit n Predicts the performance improvement n Measures improved culling efficiency when restructuring sub-BVHs

15 Major Observation l Restructuring two nodes with BV overlaps can improve the culling efficiency n Assumes that restructuring operation will remove all the BV overlaps A BVH

16 Selective Restructuring Operations

17 Overall Framework l At a new frame n Refits BVs with deformed triangles n Performs our selective restructuring algorithm n Runs BVH-based ray tracing

18 Detecting BV Overlaps l Brute-force method n Requires O(m 2 ) where m is # of BVs l Hierarchical traversal and culling n Inspired by efficient collision detection methods

19 Overview of Selective Restructuring Algorithm l Computes restructuring candidates n Detects nodes with BV overlaps during hierarchy traversal l Restructure node pairs with higher restructuring benefits greedily n Improves the performance of ray tracing

20 Evaluating Our Algorithm l Implement BVH-based ray tracer [Lauterbach et al. 06] n Tests with four dynamic scenes having different characteristics

21 Dynamic Scenes l Cloth simulation (92K)

22 Dynamic Scenes l N-body simulation (146K)

23 Dynamic Scenes l Exploding dragon (252K) l BART (65K)

24 Prior Works l BV Refitting [Wald et al. 07, Bergen 97] l Complete re-construction from scratch l RT-Deform [Lauterbach et al. 06] l LM method [Larsson and Akenine-Möller 06]

25 Performance Improvement Ratio Complete re-construction Exploding dragon 8.5 N-body simulation 1.8 BART1.1 Cloth simulation 4.7 Refitting only 11 > 80 28 0.96

26 Image Shots from Cloth Simulation

27 Performance Improvement Ratio Complete const. Refitting only Exploding dragon 8.511 N-body simulation 1.8> 80 BART 1.128 Cloth simulation 4.70.96 Robust performance improvement across our benchmarks RT- Deform LM method 1.652.16 1.251.36 2.51.11 1.031.29

28 Conclusions l Novel algorithm to selectively restructure BVHs n Based on selective restructuring operations and two BVH metrics n Has more robustness and deals with bigger scene complexity n Can be used in other applications l Dynamic scenes are available

29 Acknowledgements l Naga Govindaraju l Christian Lauterbach l Ingo Wald l Ming Jang l Hanan Samet l Peter Lindstrom l Other members of data analysis group l Anonymous reviewers l Our funding agencies


Download ppt "Ray Tracing Dynamic Scenes using Selective Restructuring Sung-eui Yoon Sean Curtis Dinesh Manocha Univ. of North Carolina at Chapel Hill Lawrence Livermore."

Similar presentations


Ads by Google