Jump Restore Light Transport

Sascha Holl,Hans-Peter Seidel,Gurprit Singh

2024-10-25

Abstract:Markov chain Monte Carlo (MCMC) algorithms come to rescue when sampling from a complex, high-dimensional distribution by a conventional method is intractable. Even though MCMC is a powerful tool, it is also hard to control and tune in practice. Simultaneously achieving both local exploration of the state space and global discovery of the target distribution is a challenging task. In this work, we present a MCMC formulation that subsumes all existing MCMC samplers employed in rendering. We then present a novel framework for adjusting an arbitrary Markov chain, making it exhibit invariance with respect to a specified target distribution. To showcase the potential of the proposed framework, we focus on a first simple application in light transport simulation. As a by-product, we introduce continuous-time MCMC sampling to the computer graphics community. We show how any existing MCMC-based light transport algorithm can be embedded into our framework. We empirically and theoretically prove that this embedding is superior to running the standalone algorithm. In fact, our approach will convert any existing algorithm into a highly parallelizable variant with shorter running time, smaller error and less variance.

Graphics,Probability

What problem does this paper attempt to address?

The paper attempts to address some limitations of existing Markov Chain Monte Carlo (MCMC) methods in light transport simulation in computer graphics. Specifically, these methods are inefficient in exploring high-dimensional state spaces, prone to getting stuck in local modes, and have slow convergence rates. These issues limit their performance in generating high-quality, realistic images.

The authors propose a new framework aimed at overcoming the aforementioned problems. This framework allows for the use of arbitrary Markov chains for Monte Carlo sampling and improves the algorithm's parallelism and efficiency by introducing continuous-time dynamics and a global discovery mechanism. Specifically, the framework:

1. **Allows the use of arbitrary Markov chains**: Users can define Markov chains with specific local exploration properties without worrying about whether they satisfy the target distribution's invariance.
2. **Continuous-time dynamics**: The algorithm is rejection-free, and all generated samples are used for state exploration and global discovery.
3. **Global discovery mechanism**: By initiating new instances of Markov chains at different locations, it ensures the exploration of the global state space.
4. **Parallel execution**: If the starting position of the next instance does not depend on the exit point of the previous instance, all processes can be executed in parallel, in stark contrast to the traditional strictly sequential Metropolis-Hastings algorithm.

With these improvements, the authors hope to provide a faster and more accurate method for light transport simulation, thereby enhancing the efficiency and quality of generating realistic images.