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.

翻译：我们提出可微分间接寻址——一种新颖的学习型原语，利用可微分多尺度查找表有效替代传统图形管线中的计算与数据操作。我们在一系列图形任务中验证了其灵活性，包括几何与图像表示、纹理映射、着色及辐射场表示。在所有场景下，可微分间接寻址均能无缝集成至现有架构，实现快速训练，并同时产出灵活高效的结果。