Deep Rendering

DisneyResearch|Studios leads developments of solutions for accelerating production rendering via machine learning with projects such as production-ready denoising, deep scattering in atmospheric clouds, or path guiding for efficient simulation of light paths.

Abstract:

Physically accurate rendering of 3D worlds requires sophisticated Monte Carlo algorithms for simulating the propagation of light. Rendering our characters in magic environments created at our studios requires immense computational resources. Disney Research focuses on developing algorithms for speeding up the rendering process and providing artists with tools for efficient asset design and fast iterations.

We leverage machine learning, and deep learning in particular, to accelerate image synthesis and simulations of light transport. We aim at significantly improving the efficiency of the core aspects of rendering, such as importance sampling and image reconstruction, with emphasis on seamless integration into current and future production pipelines.

Tech Transfer:

Disney researchers assist with the deployment of our tools in production. Independent of whether it is a specialized techniques for learning and simulating light transport, or a complete solution, such as our deep-learning filters for removing noise from rendered images, we help Pixar and Disney Animation with integrating our ideas into RenderMan, Hyperion, and studios’ production pipelines.

Publication Highlights