Light Transport Simulation
in the Gradient Domain

Despite the wide adoption in film production and animation industry nowadays, Monte Carlo light transport simulation is still prone to producing noisy images within short rendering time. Accelerating the convergence of Monte Carlo rendering without sacrificing its accuracy is by far a challenging task. In this course, we will learn about gradient-domain light transport simulation, a recent family of techniques in physically based rendering introduced in the past five years that can accelerate traditional Monte Carlo rendering up to approximately an order of magnitude based on gradient estimation and image reconstruction. Particularly, we will introduce the fundamentals of gradient-domain rendering with gradient-domain path tracing, and then extend the discussion to gradient-domain bidirectional path tracing and photon density estimation. We also discuss volume rendering in the gradient domain before diving into advanced topics in recent state-of-the-art papers in this direction. We further discuss tips and tricks in open-source implementations of such algorithms, and provide ideas for future research directions.


Tuesday, December 4, 2018
9:00am - 12:45pm
Hall D1, Tokyo International Forum


9:00am: Opening

Welcome speech and a brief introduction about the course, its objectives and schedule.

9:10am: Basic theories in light transport

We give a brief overview about the state-of-the-art Monte Carlo light transport simulation techniques and highlighted path tracing and photon mapping. This is a warming up section for those who are not very familiar with rendering before we dive into the technical details of gradient-domain rendering.

9:30am: Fundamentals of gradient-domain light transport

The key ideas of gradient-domain rendering is introduced. Particularly, the audience will learn the concepts of the modified rendering equation for gradient-domain rendering, the shift mapping function and its Jacobian. Gradient-domain path tracing will be introduced along with the simple half-vector shift mapping. The classical reconstruction technique using screened Poisson reconstruction will also be introduced.

Break (10 minutes)

10:10am: Gradient-domain bidirectional light transport

We continue with the discussion of bidirectional path tracing, progressive photon mapping, and their combinations in the gradient domain. In this part, more complex shift mapping techniques will be introduced such as manifold exploration.

10:40am: Basic theories of volumetric rendering

A brief introduction to volumetric rendering.

10:55am: Gradient-domain volumetric rendering

Beyond surfaces, we discuss volumetric rendering with the gradient-domain volumetric photon density estimation techniques. More complex concepts in volumetric rendering such as photon points, beams, and planes are introduced in this section.

Break (10 minutes)

11:20am: Advanced topics in gradient-domain rendering

We discuss a collection of many advanced techniques in gradient-domain rendering such as temporal rendering, adaptive rendering, Metropolis light transport, path reuse, spectral rendering, and improved reconstruction.

11:50am: Practical tips and tricks of implementing gradient-domain rendering

We discuss practical issues and solutions for implementing gradient-domain rendering in Mitsuba, an open-source physically based renderer. We also share our experience in implementing a hobby-time gradient-domain renderer from scratch in Rust, a system programming language designed to prevent segmentation faults.

12:05pm: Conclusions

We will summarize the key concepts we learn so far, and brainstorm ideas for future work.


Binh-Son Hua

The University of Tokyo

Adrien Gruson

The University of Tokyo

Matthias Zwicker

University of Maryland

Toshiya Hachisuka

The University of Tokyo