A new importance-sampling scheme estimates Hessians and Hessian-vector products of smoothed renderings, enabling Newton-type optimizers in inverse rendering with O(1) rendering evaluations per step.
Transform- ing a non-differentiable rasterizer into a differentiable one with stochastic gradient estimation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.GR 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Stochastic Gradient Estimation for Higher-order Differentiable Rendering
A new importance-sampling scheme estimates Hessians and Hessian-vector products of smoothed renderings, enabling Newton-type optimizers in inverse rendering with O(1) rendering evaluations per step.