User:Samkottler/GSoC2020/Proposal

Name
Sam Kottler

Contact
Email: [mailto:dev@samkottler.net dev@samkottler.net] Github: https://github.com/samkottler blender.chat: samkottler

Synopsis
This project will continue development on the Many Light Sampling project from GSoC 2018. It will add support for importance sampling of more features and make that project production ready by stabilizing the existing implementation. The goal of the original Many Light Sampling project was to use multiple importance sampling to avoid having to sample every light source directly for every shading point. The implantation is based on a paper [1] and makes efficient use of a BVH data structure to choose which light source(s) a shading point should sample.

Benefits
Using multiple importance sampling to selectively choose which lights to sample will help make Cycles more efficient at rendering environments with many light sources. This technique could make it possible to render scenes with millions of light sources (for example, from a mesh) within a reasonable amount of time. This will give more freedom to artists who will not need to worry about performance penalties as a result of including a large number of lights.

Deliverables
The minimum deliverable for this project would be a production ready implementation of the previous Many Light Sampling project, including volume support and extensive testing. This will mostly be adding support for the algorithm described in [1] to Cycles. It will also include a few user-facing settings such as a checkbox to enable this feature and a slider to control the splitting threshold. Finally, this project will provide end-user documentation giving a high level overview of the technique.

Project Details
This project has a number of features that will need to be completed to make the existing implementation production ready.
 * Test on production scenes and fix issues: The 2018 project was not meant to completely finish the implementation of this feature. Thus one of the most important things to complete to make this feature production ready is to test the algorithm on existing production scenes and ensure that it still gives a correct result. In the case that there are bugs, they will be fixed.
 * Complete volume rendering support: While some work for volume    rendering was done, it is still very much a work in progress and more work and testing need to be done to make it production ready.
 * Investigate importance sampling: One thing that was left incomplete during the 2018 GSoC was balancing importance with background and portal lights. Additionally, there may have been other improvements to the algorithm made in the last two years. This    project will complete this investigation and make changes to the implementation that will increase performance.
 * Complete GPU rendering support and improve importance estimation for    textured lights: these are more features that were not completed by the 2018 project which can improve performance.

Project Schedule
June 1 – June 15: Familiarize myself with the implementation from the 2018 project, study the details of [1], and research any improvements to this technique that have been made since then June 15 – June 29: Test existing implementation and fix bugs June 29 – July 27: Add support for volume rendering, including time for testing with existing production scenes July 27 – August 17: Complete GPU support and improve importance estimation for textured lights, including time for testing August 17 – August 24: Buffer time to allow for unexpected delays. If this is unneeded, this time will be used to add more features to use for importance sampling.

Bio
My name is Sam Kottler and I was a double major in math and computer science at Colorado College, where I am currently the computer science paraprofessional. In this position I run problem sessions for courses, help professors lecture, and have time to work on my own projects. I will be attending the computer science PhD program at the University of Wisconsin at Madison starting in the fall. During my studies, I have worked on multiple 3D rendering and animation projects, including currently working through the book Physically Based Rendering. My background in math has made it easier for me pick up rendering algorithms such as Path Tracing and Metropolis-Hastings sampling. I have also worked on a few GPU rendering projects in both CUDA and OpenCL, which can be found on my github linked above.