pouët.net

Go to bottom

Separable Subsurface Scattering from EGSR 2015

category: code [glöplog]
 
http://cg.tuwien.ac.at/~zsolnai/gfx/separable-subsurface-scattering-with-activision-blizzard/

also...lets do it in 4k :P
At least link it correctly..
added on the 2015-05-12 19:33:03 by Hofstee Hofstee
Also heck, it's already 4k.
added on the 2015-05-12 19:33:48 by Hofstee Hofstee
Of course it's implemented in 4k. It's only a proof of concept though. Now what's left is to make an actual 4k intro out of it. ;)
added on the 2015-05-13 00:15:44 by kzf kzf
Well done, another S concatenated. Next one please. No combo breaker please
How about the Prefixing Screen-Space ;)
it looks like all it is is an additional lighting pass that lights shadowed fragments with a term that's attenuated using the "thickness" (from the light's point of view using shadow map), and then blurs (/bilaterally filters) the shit out of it with a slight shift towards red in the kernel?
(or did i miss the point completely)

those kinds of graphics paper are the ones i kindof loathe: a relatively simple and heavily faked technique, upsold by being made to sound incredibly complex, and justified with loads of maths to prove how the fake's actually really ok honest. the kind of things non-acaedemia would put in a one-page blog post.

fine example of the importance of good art assets in "selling" something, though.
added on the 2015-05-13 15:40:00 by smash smash
Quote:

it looks like all it is is an additional lighting pass that lights shadowed fragments with a term that's attenuated using the "thickness" (from the light's point of view using shadow map), and then blurs (/bilaterally filters) the shit out of it with a slight shift towards red in the kernel?
(or did i miss the point completely)

those kinds of graphics paper are the ones i kindof loathe: a relatively simple and heavily faked technique, upsold by being made to sound incredibly complex, and justified with loads of maths to prove how the fake's actually really ok honest. the kind of things non-acaedemia would put in a one-page blog post.

fine example of the importance of good art assets in "selling" something, though.

Well, it seems to work quite okayish. Didn't try it yet. Would expect too much from doing some blurs.

IMHO academic papers are most of the time a bad representation for something implemented in code.
Often a well documented code sample and a 60 min video of someone walking you through the important details would be much better.
One of the main problems is that nobody is really honest about measurements and details nowadays, that creates some pressure to do the same... Downwards spiral...

This publication is doing it far better than most of the other ones, there are shitloads of supplemental material, even a 4k implementation. You can easily try it yourself (that's awesome) - you can decide whether it works for your use case or not, without going through a complete implementation hell).
Releasing code is "dangerous", since you can not fake your shit anymore (people do that and it's much worse than you think - I have seen things I can not unsee) and if you screwed something up, others can figure it out.

Besides the selling stuff and the art assets which clearly help to sell it - this publication is doing a lot of things right and I have some respect for that without diving into the code.

In this context, maybe interesting: "Why not"

Conferences should ENFORCE authors to release code. Everything else is just killing resources and a waste of time. My 50 cent.
added on the 2015-05-13 16:14:10 by las las
las: i agree with you, this paper is doing a lot better than most because it actually releases the code and a demo you can manipulate.
but in terms of the paper, well.. it'd have been better without the paper.

i guess all of academia probably suffers in some respects from bloated papers and overselling and so on, but in the graphics business it really feels out of place: you quite often see techniques you'd come up with in a morning to fake some effect, full of pitfalls and limitations, but written up into a 5 page paper full of equations to prove how correct it is. perhaps its because there's no gap, or even a negative gap between the academics & industry (and even hobbyists), it makes the dubiousness of it much more apparent.
added on the 2015-05-13 17:03:58 by smash smash
Guys, I know that I will mostly state the obvious, but anyway, some papers in computer graphics field are closely related to implementation/hardware - like this one, and are trying to show that their implementation is efficient on a current hardware etc... so obviously releasing a code makes sense. On the other hand, some papers are more theoretical in nature, where formulas itself are interesting enough and implementation is just a "technical detail", so enforcing the code in this case would be ironically - another kind of bloat.

About this particular publication: I think they did pretty good job, also nice touch with 4k implementation, and actually I don't mind 5 pages, in fact it's better to have a little more extensive study (that you can mostly skip if you don't care) than just few slides with imprecise bullet-based language you would normally get from the so called industry ;)
added on the 2015-05-14 02:03:03 by tomkh tomkh
Paper has no pseudocode just matrices and integrals in formulas..
c++ code is too long <delete>.
added on the 2015-05-14 02:28:45 by rudi rudi

login

Go to top