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.
%0 Conference Paper
%1 https://doi.org/10.48550/arxiv.2204.13845
%A Petersen, Felix
%A Goldluecke, Bastian
%A Borgelt, Christian
%A Deussen, Oliver
%B Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR)
%D 2022
%K 2022 a04 b05 from:christinawarren sfbtrr161
%P 3992-4001
%R 10.1109/CVPR52688.2022.00397
%T GenDR: A Generalized Differentiable Renderer
%U https://doi.ieeecomputersociety.org/10.1109/CVPR52688.2022.00397
%X 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.
@inproceedings{https://doi.org/10.48550/arxiv.2204.13845,
abstract = {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.},
added-at = {2022-09-23T09:22:01.000+0200},
author = {Petersen, Felix and Goldluecke, Bastian and Borgelt, Christian and Deussen, Oliver},
biburl = {https://puma.ub.uni-stuttgart.de/bibtex/23905452c8d75fe80376b3c2cd5ca4926/sfbtrr161},
booktitle = {Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR)},
copyright = {arXiv.org perpetual, non-exclusive license},
doi = {10.1109/CVPR52688.2022.00397},
interhash = {46e3ea0fb799bef49bf005ab80794c54},
intrahash = {3905452c8d75fe80376b3c2cd5ca4926},
keywords = {2022 a04 b05 from:christinawarren sfbtrr161},
pages = {3992-4001},
timestamp = {2023-02-23T09:17:00.000+0100},
title = {GenDR: A Generalized Differentiable Renderer},
url = {https://doi.ieeecomputersociety.org/10.1109/CVPR52688.2022.00397},
year = 2022
}
