Efficient Graphics Representation with Differentiable Indirection

TL;DR

This paper presents differentiable indirection, a novel learned primitive using multi-scale lookup tables that enhances various graphics tasks by providing a flexible, efficient, and easily integrable alternative to traditional methods.

Contribution

Introduction of differentiable indirection, a new primitive employing multi-scale lookup tables for improved graphics representations and processing.

Findings

Seamless integration into existing graphics architectures.

Rapid training and versatile application across tasks.

Achieves efficient and high-quality results in graphics tasks.

Abstract

We introduce differentiable indirection -- a novel learned primitive that employs differentiable multi-scale lookup tables as an effective substitute for traditional compute and data operations across the graphics pipeline. We demonstrate its flexibility on a number of graphics tasks, i.e., geometric and image representation, texture mapping, shading, and radiance field representation. In all cases, differentiable indirection seamlessly integrates into existing architectures, trains rapidly, and yields both versatile and efficient results.