Markov Chains for Light Transport
Event details
| Date | 29.08.2018 |
| Hour | 10:00 › 12:00 |
| Speaker | Merlin Nimier-David |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Mark Pauly
Thesis advisor: Prof. Wenzel Jakob
Co-examiner: Prof. Sabine Süsstrunk
Abstract
Physically Based Rendering presents important computational and algorithmic challenges. Accurate simulation of light transport involves not only appearance models that capture real-world behavior, but also requires powerful numerical integration techniques. However, the most popular algorithm, Path Tracing, struggles with many lighting scenarios and generates incoherent workloads, which make poor use of modern hardware capabilities such as vectorization.
We summarize and compare three previous works, first recalling the background most relevant to these techniques. We then cover the three selected publications: "Riemannian Manifold Hamiltonian Monte Carlo" [Girolami et al. 2009], "A Simple and Robust Mutation Strategy for the Metropolis Light Transport Algorithm" [Kelemen et al. 2002], and "Anisotropic Gaussian Mutations for Metropolis Light Transport Through Hessian-hamiltonian Dynamics" [Li et al. 2015].
We conclude by outlining our ongoing research. Related to the three methods presented, it makes use of Markov Chain Monte Carlo integration and defines an advanced mutation strategy, which is designed to generate coherent workloads. Finally, we address future work and general research directions.
Background papers
A simple and robust mutation strategy for the metropolis light transport algorithm, by Kelemen, Csaba, et al. Computer Graphics Forum. Vol. 21. No. 3. Blackwell Publishing, Inc, 2002.
Anisotropic gaussian mutations for metropolis light transport through hessian-hamiltonian dynamics, by Li, Tzu-Mao, et al. ACM Transactions on Graphics (TOG) 34.6 (2015): 209.
Riemannian Manifold Hamiltonian Monte Carlo, by Girolami, Mark, Ben Calderhead, and Siu A. Chin. arXiv preprint arXiv:0907.1100 (2009). [excluding sections 6 and 7] .
Exam president: Prof. Mark Pauly
Thesis advisor: Prof. Wenzel Jakob
Co-examiner: Prof. Sabine Süsstrunk
Abstract
Physically Based Rendering presents important computational and algorithmic challenges. Accurate simulation of light transport involves not only appearance models that capture real-world behavior, but also requires powerful numerical integration techniques. However, the most popular algorithm, Path Tracing, struggles with many lighting scenarios and generates incoherent workloads, which make poor use of modern hardware capabilities such as vectorization.
We summarize and compare three previous works, first recalling the background most relevant to these techniques. We then cover the three selected publications: "Riemannian Manifold Hamiltonian Monte Carlo" [Girolami et al. 2009], "A Simple and Robust Mutation Strategy for the Metropolis Light Transport Algorithm" [Kelemen et al. 2002], and "Anisotropic Gaussian Mutations for Metropolis Light Transport Through Hessian-hamiltonian Dynamics" [Li et al. 2015].
We conclude by outlining our ongoing research. Related to the three methods presented, it makes use of Markov Chain Monte Carlo integration and defines an advanced mutation strategy, which is designed to generate coherent workloads. Finally, we address future work and general research directions.
Background papers
A simple and robust mutation strategy for the metropolis light transport algorithm, by Kelemen, Csaba, et al. Computer Graphics Forum. Vol. 21. No. 3. Blackwell Publishing, Inc, 2002.
Anisotropic gaussian mutations for metropolis light transport through hessian-hamiltonian dynamics, by Li, Tzu-Mao, et al. ACM Transactions on Graphics (TOG) 34.6 (2015): 209.
Riemannian Manifold Hamiltonian Monte Carlo, by Girolami, Mark, Ben Calderhead, and Siu A. Chin. arXiv preprint arXiv:0907.1100 (2009). [excluding sections 6 and 7] .
Practical information
- General public
- Free
Contact
- EDIC - [email protected]