160 likes | 259 Views
SIGGRAPH 2011 ASIA Preview Seminar Rendering: Accuracy and Efficiency. Shinichi Yamashita Triaxis Co.,Ltd. Rendering: Accuracy and Efficiency Paper List.
E N D
SIGGRAPH 2011 ASIA Preview SeminarRendering: Accuracy and Efficiency Shinichi Yamashita Triaxis Co.,Ltd.
Rendering: Accuracy and EfficiencyPaper List • Displacement Interpolation Using Lagrandian Mass Transport Nicolas Bonneel et al. (National Institute for Research in Computer Science and Control, Canada) • Adaptive Sampling and Reconstruction using Greedy Error MinimizationFabrice Rousselle et al. (University of Bern, Switzerland) • T&I Engine: Traversal and Intersection Engine for Hardware Accelerated Ray TracingJae-Ho Nah et al. (Yonsei University, South Korea) • Coherent Parallel HashingIsmael Garcia Fernandez et al. (University of Girona, Spain)
Displacement Interpolation Using Lagrandian Mass TransportAbstract 1 • Linear Interpolation: can’t capture transitional motion • Displacement Interpolation: works based on advection • Distributions or functions are decomposed into sum of radial basis functions (RBFs). • Algorithm finds pair of RBFs and compute their mass transport to obtain the interpolated function.
1. Displacement Interpolation Using Lagrandian Mass TransportDisplacement Interpolation If we know the distribution function, displacement interpolation is easy to be calculated. But how we could interpolate distributions whose parameterized formula is unknown?
1. Displacement Interpolation Using Lagrandian Mass TransportAlgorithm Refer to the Paper Video
1. Displacement Interpolation Using Lagrandian Mass TransportResults and Applications Also Refer to the Paper Video
1. Displacement Interpolation Using Lagrandian Mass TransportConclusion • Displacement interpolation is useful in many CG situations rather than linear interpolation. • General approach of displacement interpolation on multi-dimensional and continuous domain is proposed. • Several applications of displacement interpolation is demonstrated.
Adaptive Sampling and Reconstruction using Greedy Error MinimizationAbstract 2 • A new adaptive sampling method for Monte Carlo rendering. • It operates in image space and uses iterative approach. • It focuses on minimizing the MSE on same # of samples.
2. Adaptive Sampling and Reconstruction using Greedy Error MinimizationAlgorithm Overview Rendered image Per-pixel Filters Output Image Monte Carlo renderer Filter Selection Algorithm New samples Note: each algorithm gives solution which minimize the estimated MSE on output image. Sample Distribution Algorithm
2. Adaptive Sampling and Reconstruction using Greedy Error MinimizationResult
2. Adaptive Sampling and Reconstruction using Greedy Error MinimizationConclusions • An adaptive sampling and reconstruction algorithm that greedily minimize MSE in Monte Carlo rendering is described. • It significantly improves MSE and image quality over previous work. • A robust filter selection algorithm.
T&I Engine: Traversal and Intersection Engine for Hardware Accelerated Ray TracingAbstract 3 • A new ray tracing hardware architecture is proposed • It integrates 3 novel architectures; • cache-efficient layout and its traversal • 3-phase ray-triangle intersection test • latency hiding defined as the ray accumulation unit Paper is not available
Coherent Parallel Hashing Abstract 4 • A new spatial hashing algorithm for parallel GPU processing. • High load factor and quick rejection time for empty key query. • Preserves space coherence so that adjacent data is kept together in hash table.
4. Coherent Parallel Hashing Algorithm • Based on Robin Hood hashing [Celis 1986]. • Store maximum age of key with each entry in hash table for quick rejection of empty key. • Probe function is defined as follows to keep spatial coherence.
4. Coherent Parallel Hashing Examples (Texture Painting) Atras:4096x4096 1M pixels are painted. Hash construction time and rendering frame rate are 3ms and 446fps for d=0.5, 10ms and 237fps for d=0.99.
4. Coherent Parallel Hashing Conclusions • A new spatial hashing method is introduced. • It can be constructed by GPU in parallel. • It keeps high load factor and fast query time (nature of Cuckoo hashing [Pagh and Rodler 2004]). • Empty key queries are quickly rejected. • Spatial coherence is maintained. • Implemented on nVIDIA Fermi GTX480 (CUDA) • It showed remarkable performance.