Systematically Differentiating Parametric Discontinuities

ACM SIGGRAPH 2021

Jesse Michel*
MIT CSAIL
Kevin Mu
MIT CSAIL
Gilbert Bernstein
UC Berkeley and
MIT CSAIL
Tzu-Mao Li
MIT CSAIL
 
overview

Abstract

Emerging research in computer graphics, inverse problems, and machine learning requires us to differentiate and optimize parametric discontinuities. These discontinuities appear in object boundaries, occlusion, contact, and sudden change over time. In many domains, such as rendering and physics simulation, we differentiate the parameters of models that are expressed as integrals over discontinuous functions. Ignoring the discontinuities during differentiation often has a significant impact on the optimization process. Previous approaches either apply specialized hand-derived solutions, smooth out the discontinuities, or rely on incorrect automatic differentiation.  

We propose a systematic approach to differentiating integrals with discontinuous integrands, by developing a new differentiable programming language. We introduce integration as a language primitive and account for the Dirac delta contribution from differentiating parametric discontinuities in the integrand. We formally define the language semantics and prove the correctness and closure under the differentiation, allowing the generation of gradients and higher-order derivatives. We also build a system, Teg, implementing these semantics. Our approach is widely applicable to a variety of tasks, including image stylization, fitting shader parameters, trajectory optimization, and optimizing physical designs.

Short Summary

Full Video Presentation

BibTeX

Acknowledgements

We thank Fredo Durand for the discussion of the idea in the early stage and proofreading, Ante Qu for tips on handling friction, Paul Zhang for his discussions on image triangulation and diffeomorphisms, Joshua Fishman and Tao Du for their advice on physical simulation methods, Samuel Tenka for his insightful discussions of distribution theory, and Luke Anderson for his detailed proof-reading. This research was funded under DARPA agreement HR00112090017.