Progressive Photon Beams back

(L) [2011/09/26] [jufuny8138] [Progressive Photon Beams] Wayback!

Worjciech Jarosz has recently released the photon beam-'ver2'.
looking practical in implementation if you have beam photon mapper.
download [LINK http://zurich.disneyresearch.com/~wjarosz/publications/jarosz11progressive.pdf]

By the way,,, anybody can guess better word other than ('adaptive' || 'progressive' || 'robust' || 'efficient' ) && GPU...?  [SMILEY 8)]  anyway "progressive" is still an very nice and trendy vocabulary ... [SMILEY :)]

(L) [2011/09/26] [ingenious] [Progressive Photon Beams] Wayback!

I'd also recommend watching the video on their web page:
[LINK http://zurich.disneyresearch.com/~wjarosz/publications/jarosz11progressive.html]
It's not... all bad [SMILEY 8)] As for the words - they all have different meaning actually. You can come up with an algortihm that is, say, any two of the above but not the other two [SMILEY :)]

(L) [2011/09/29] [thachisu] [Progressive Photon Beams] Wayback!

I may sound annoying to bring up this point, but I am wondering how their approach compares with BRE + SPPM that I touched briefly in this post:
[LINK http://ompf.org/forum/viewtopic.php?f=10&t=1492&start=150#p23427 viewtopic.php?f=10&t=1492&start=150#p23427]
It does converge to the correct solution too, but I would imagine that the use of beams is important for performance.
In any case, it is great to see that progressive radiance estimate is not incompatible with beams! At the same time, this subject is getting crowded so I'd better find another toy :->

(L) [2011/09/29] [ingenious] [Progressive Photon Beams] Wayback!

I haven't read the beams papers in detail yet, but it actually makes quite a bit of sense to use beams in the volumes for computational efficiency reasons. Essentially, with volumetric photon mapping, if you have a ray piercing the volume and distributing energy in it, you represent this whole line with a randomly chosen point somewhere along it. And now if you have another, viewing ray, with some associated width/beam, that wants to integrate radiance in the volume, what's the chance that this randomly chosen point is inside that beam? And for the next viewing ray? That's my intuition. Storing the whole line directly can result in massive variance decrease. The line/photon beam radiance estimate might be more costly now, but apparently you need an order of magnitude fewer lines for the same variance levels. That's all in Wojciek's TOG paper on photon beams, which I'm sure you're aware of [SMILEY :)] Now, it's interesting how efficient these beams are in inhomogeneous volumes. Again, they might be, haven't had the time to read either paper.
What I really liked in the progressive beams video is the structure of the noise of this new algorithm [SMILEY :)] No splotches anymore, but some lines wiggling around [SMILEY :)]