“Neural Parametric Mixtures for Path Guiding” by Dong, Wang and Li
Conference:
Type(s):
Title:
- Neural Parametric Mixtures for Path Guiding
Session/Category Title: Neural Light Transport
Presenter(s)/Author(s):
Moderator(s):
Abstract:
We present Neural Parametric Mixtures (NPM), a neural formulation to encode target distributions for path guiding. NPM uses a continuous and compact representation for encoding parametric distributions and decodes them via lightweight MLPs. Gradient-based optimization strategy can efficiently train NPM while being practical for parallelization on GPUs.