Neural Path Guiding with Distribution Factorization

Pedro Figueiredo, Qihao He, and Nima Khademi Kalantari

Evaluation

In this supplementary visualizer, we compare the two variants of our approach (DF-N and DF-L) against unidirectional path tracing and four state-of-the-art methods: Müller et al. [2017] (PPG), Rath et al. [2020] (Variance), Müller et al. [2019] (NIS), Dong et al. [2023] (NPM). We report three metrics: ꟻLIP (Andersson et al. [2020]), mean absolute percentage error (MAPE), and relative mean squared error (relMSE). relMSE is computed by discarding the top 0.1% errors to alleviate the contribution of outliers.

*The errors reported in this visualizer are averaged over ten runs, same as presented in Table 1. The images are from just one run.