Learning to Cluster for Rendering with Many Lights

Yu-Chen Wang1     Yu-Ting Wu1     Tzu-Mao Li2,3     Yung-Yu Chuang1
National Taiwan University1     MIT CSAIL2     University of California San Diego3    

All results are generated on a machine with 8-core Intel Core i7-9700 CPU (using 4 cores) and 32GB RAM.
Image viewer is borrowed from this page.

Scene (click on the thumbnail to select the scene comparison shown at the bottom of the page)
Bathroom
(120 sec.)
Bedroom
(120 sec.)
Classroom
(120 sec.)
Kitchen with VPL.
(120 sec.)
Living-room
(60 sec.)
Parking-lot
(360 sec.)
Sanmiguel with VPL.
(480 sec.)
SiA-shelf
(30 sec.)
Staircase
(30 sec.)
Staircase2
(60 sec.)

Please select mode:

Result Comparison       Ablation Study

Compared methods:
 Stochastic Lightcuts (SLC) [Cem Yuksel 2019]
Resampled Importance Sampling (RIS) [Talbot et al. 2005, Bitterli et al. 2020]
Bayesian Online Regression for Adaptive Direct Illumination Sampling (BORAS) [Vevoda et al. 2018]
Variance-aware BORAS (VA-BORAS) [Rath et al. 2020, Vevoda et al. 2018]
Importance Sampling of Many Lights with Reinforcement Lightcuts Learning (RLL) [Pantaleoni 2019]