Support NeoGAF

[Durante]

I'm asking myself if CBR can't also be used for speeding up some global
illumination? For, let's say we cover the sphere (over which we will
integrate) with a set of well chosen points (not random) and compute the
illumination over half of these points (in a checkerboard fashion) while
backward integrating the other half from a previous frame, like in classic
CBR, yet all done on the sphere, then we would only need half the rays saving
about half the computation. Anyone?

Well, screen space ambient occlusion is an approximation of global illumination, usually casts rays in a half-sphere, and IIRC Ubisoft already temporally re-used samples for that back in AC:Unity.
Of course, that was in screen space, but I think the basic idea is quite similar to what you are proposing.

 

[missile]

Well, screen space ambient occlusion is an approximation of global illumination, usually casts rays in a half-sphere, and IIRC Ubisoft already temporally re-used samples for that back in AC:Unity.
Of course, that was in screen space, but I think the basic idea is quite similar to what you are proposing.

Yeah, could be the case. I guess most of these techniques boil down to the
same principles, trying to hide one class of artifacts better than another
one.

I will give CBR a try if I have my GI integrator running.

 

[Fafalada]

It shades half as many samples per frame, that's a rather simple statement. And, more importantly, accurate.

That still depends on how exactly we define "shading" - typically you will want as much of post to run post-reconstruction as possible to maximize quality, and where exactly you draw the line between "shading" and "not shading" can get kind of fuzzy, especially if you're dealing with a deferred renderer to begin with.
I guess most typical definition would be "light contribution pass" but even that isn't guaranteed to entirely fall on the pre-reconstruction side - of course, all very app-dependant in the end.

I will give CBR a try if I have my GI integrator running.

If you're looking at reducing runtime costs of realtime diffuse contributions - I'd prefer to look into texture-space/decoupled shading over sample-reduction schemes. Obviously, doesn't map that well to view-dependent contributions though.

 

[missile]

... If you're looking at reducing runtime costs of realtime diffuse contributions - I'd prefer to look into texture-space/decoupled shading over sample-reduction schemes. Obviously, doesn't map that well to view-dependent contributions though.

View-dependent or die. xD Nah, I'm just looking for some different techniques
to play around with.

 

[missile]

How does CBR behave for transparent objects? Does it hold up as well?

 

[pottuvoi]

Didn't insomniac also used temporal data to render shadow maps?

Seems like it
Cascaded shadow map scrolling

That is pretty much identical how old good 2D games handled scrolling.

Background layer with no moving parts and sprite layer which handles characters etc.
Background is calculated in tiles at edge areas outside 'screen' when shadowmap projector moves.
Both maps are combined to get final result for masking light.

At least some of the shading could be handled in similar way as well, detached from resolution or moving objects. (View dependant light is problematic though.)
A return to surface caches, texture based shading, tracking shaded samples world space.. etc.
http://gpuopen.com/texel-shading/
https://forum.beyond3d.com/threads/object-texture-space-lighting.57676/
https://www.bluesnews.com/abrash/chap68.shtml

How does CBR behave for transparent objects? Does it hold up as well?

Really depends how transparent surfaces are rendered and if they go through CBR at all.

 

[KyleOnTheRun]

Freakin' A. Fantastic thread and I feel like I legitimately understand what this technique means. Plus, *loved* the visual GIFs demonstrating differences between the three.

I didn't even know that pixels approximate the polygons "beneath" them, so I can safely say I learned a ton.

 

[PLASTICA-MAN]

This reddit thread used this thread as reference and added some other interesting infos.

 

[ValloaYasu]

RedGamingTech just realised a video about "Frostbite Dev Details Performance Gains on Checkerboard Rendering"

Some nice infos

https://youtu.be/fqFQn39yuhM

 

[Spladam]

Impressive breakdown OP, thank you very much. Did you make the example slides yourself? If so, what did you use if you don't mind me asking.

 

[Liabe Brave]

Impressive breakdown OP, thank you very much. Did you make the example slides yourself? If so, what did you use if you don't mind me asking.

Sorry for the delay responding, I haven't checked this thread in a while. Yes, all the graphics were made by me. I used GIMP 2.8 to do them. This was the first time I'd ever used that program. I'm used to Photoshop, but since I upgraded my PC I don't have a copy (and I can't afford to buy a newer version right now). Coming from more familiar territory, I found GIMP to have some very frustrating default behavior that I had to squash with lots of preference changes. But in the end it worked out fine.

 

[onQ123]

Bumped to help move the CBR talk out of the other threads

 

[FatKingBallman]

Will bump this thread for a good reason.

 

[MasterCornholio]

Nice necrobump and alot of this is still relevant today. It's a very interesting explanation on how CBR works.