GenDR: A Generalized Differentiable Renderer
This work provides a framework for differentiable rendering, which is incremental as it builds on existing methods like SoftRas and DIB-R to improve 3D reconstruction tasks.
The authors tackled the problem of designing a generalized differentiable renderer by formalizing component requirements and instantiating it with various smoothing distributions, finding that a simple uniform distribution yields the best overall results on the ShapeNet benchmark, averaging over 13 classes.
In this work, we present and study a generalized family of differentiable renderers. We discuss from scratch which components are necessary for differentiable rendering and formalize the requirements for each component. We instantiate our general differentiable renderer, which generalizes existing differentiable renderers like SoftRas and DIB-R, with an array of different smoothing distributions to cover a large spectrum of reasonable settings. We evaluate an array of differentiable renderer instantiations on the popular ShapeNet 3D reconstruction benchmark and analyze the implications of our results. Surprisingly, the simple uniform distribution yields the best overall results when averaged over 13 classes; in general, however, the optimal choice of distribution heavily depends on the task.