A machine-learning technique for reconstructing sparsely sampled image sequences rendered using Monte Carlo methods. The primary focus is on reconstruction of global illumination with extremely low sampling budgets at interactive rates.

This paper demonstrates an approach for achieving both modularity and high performance for shader code in current and upcoming real-time rendering engines. It presents a shader compiler and an engine design that enables authoring modular shader code and maps that to efficient a parameter-binding model on modern graphics APIs.