← All resources

TinyDiffRast project

Auto-discovered

Course Advanced

Jordan J. Bannister

Visit resource ↗

Description

TinyDiffRast project is a tiny course and codebase by Jordan J. Bannister, developed at Mila with Derek Nowrouzezahrai of the McGill Graphics Lab, on differentiable triangle rasterization for inverse rendering. The URL is the course landing page.

It covers the discontinuity problem, learnable cameras, rasterization and shading, automatic differentiation, optimizing lights and albedo, Rasterize Then Splat, Nvidia Diff Rast style methods, and optimizing camera and geometry parameters. Demos include interactive rendering, Jacobian images, and inverse lighting, materials, cameras, and geometry.

It is for intermediate to advanced students and researchers who know basic graphics pipelines. Linear algebra and some machine-learning familiarity help with the optimization lessons.

Access is a free GitHub Pages site with numbered lessons and an accompanying codebase.

The course is useful as a from-scratch path into inverse rendering via differentiable rasterization. Wider community reputation beyond the authors' lab affiliation is not stated on the page.