Efficient Graphics Representation with Differentiable Indirection
This addresses efficiency and flexibility issues in graphics representation for researchers and practitioners, though it appears incremental as it builds on existing architectures.
The paper tackles the problem of inefficient compute and data operations in graphics pipelines by introducing differentiable indirection, a learned primitive using multi-scale lookup tables, which integrates into existing architectures, trains rapidly, and yields versatile and efficient results across tasks like geometric representation and radiance fields.
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.