@HereForOneGhost@sensha_do@MintFantome oh does the version that's coming out on monday have anything new? or is it just trimming down the original vod? i'm debating whether to watch it now or later
@cyanoacry I've been idly thinking about this ever since pomu put herself on that high altitude weather balloon. unfortunately getting to orbit and beyond is probably still beyond the reach of a fully fan (funded + made) effort 🥲
@lisyarus Perhaps the mismatch is simply too big then? In my experiments, I found that ignoring even just one term (visibility, so V * Li * fr -> Li * fr) introduced a surprising amount of noise
@lisyarus I wasn't sure from your description whether you're using {hemisphere, Gaussian, SH, etc.} purely for sample generation, or to calculate the weights as well
@lisyarus I could be way off-base on this, but I think a key part of reservoir sampling is using a target function (weight) that matches the integrand? If any of the terms are neglected - visibility, Li, BRDF - you'll end up with some residual noise
@KostasAAA Very cool! is there some kind of "canonical reference" for the spatial hashing technique? I've seen the term used for GI solutions (like this), some ML stuff (instant-NGP), collision detection/nearest neighbor search... it's not clear to me if they're referring to the same thing