Stochastic Gradient Estimation for Higher-Order Differentiable Rendering

Abstract

We derive methods to compute higher order differentials (Hessians and Hessian-vector products) of the rendering operator. Our approach is based on importance sampling of a convolution that represents the differentials of rendering parameters and shows to be applicable to both rasterization and path tracing. We demonstrate that this information improves convergence when used in higher-order optimizers such as Newton or Conjugate Gradient relative to a gradient descent baseline in several inverse rendering tasks.

Cite

Text

Wang et al. "Stochastic Gradient Estimation for Higher-Order Differentiable Rendering." International Conference on Computer Vision, 2025.

Markdown

[Wang et al. "Stochastic Gradient Estimation for Higher-Order Differentiable Rendering." International Conference on Computer Vision, 2025.](https://mlanthology.org/iccv/2025/wang2025iccv-stochastic/)

BibTeX

@inproceedings{wang2025iccv-stochastic,
  title     = {{Stochastic Gradient Estimation for Higher-Order Differentiable Rendering}},
  author    = {Wang, Zican and Fischer, Michael and Ritschel, Tobias},
  booktitle = {International Conference on Computer Vision},
  year      = {2025},
  pages     = {28198-28206},
  url       = {https://mlanthology.org/iccv/2025/wang2025iccv-stochastic/}
}