Efficient Graphics Representation with Differentiable Indirection

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.

arXiv commentProject website: https://sayan1an.github.io/din.html

References in corpus (8)

• Deep Learning in Neural Networks: An Overview
• Instant Neural Graphics Primitives with a Multiresolution Hash Encoding
• Real-time Neural Radiance Caching for Path Tracing
• ReLU Fields: The Little Non-linearity That Could
• Random-Access Neural Compression of Material Textures
• Neural-PIL: Neural Pre-Integrated Lighting for Reflectance Decomposition
• NeuMIP: Multi-Resolution Neural Materials
• Neural Shadow Mapping