Differentiable rendering pipeline for inverse graphics

Abstract

Provided is a system, a method, and a computer readable medium for inverse graphics rendering. The system comprises a differentiable rendering pipeline and a gradient descent optimization engine. A given scene is described using scene parameters. Visibility functions, and other rendered functions, are constructed to be continuous and differentiable, allowing the optimization engine and the rendering pipeline to efficiently iterate through increasingly refined scene models.

Description

[0001] claim priority [0002] This application claims U.S. Provisional Application No. 62 / 554,943, entitled "END-TO-END DIFFERENTIABLE RENDERING PIPELINE FOR INVERSE GRAPHICS," filed September 6, 2017 (Attorney Docket No. NVIDP1187+ / 17KI0185US01), the entire contents of which are incorporated herein by reference. technical field [0003] The present invention relates to computer-generated graphics, and more particularly, to a differentiable rendering pipeline for inverse graphics. Background technique [0004] An inverse graphics system generates a three-dimensional (3D) scene model from one or more images of an observed scene. The 3D scene model may include one or more 3D object models and one or more lighting sources in the observed scene, with the goal of faithfully reproducing object model geometry and scene lighting. State-of-the-art inverse graphics systems typically perform object model generation and optimization using specialized code designed to recognize model...

AI Technical Summary

Problems solved by technology

This specialized code is usually only applicable to a specific object category, thus limiting the adaptability and versatility of prior art reverse graphics systems

Images