i think i can speed up (my implementation of) both the volumetric light methods seen here by sampling a kd-tree with photons deposited stochastically in the "air"