Illumination-Aware Spatial Subdivision for Path Guiding

We present an adaptive k-d tree subdivision method for path guiding based on variation of lighting. Our method can stop subdividing the regions with uniform illumination and can refine more aggressively for regions with high variation of lighting. Our method adapt consistently to scenes with different lighting conditions, introduce negligible memory and runtime overhead, and can replace the long standing sample-count-based subdivision used by previous path guiding works.