Large Rendering - SSAO24 and SSAO24r ????

[Raydianze]

Jesse....
* for large-rendering... will you yet be able to implement SSAO24 and SSAO24r huh?
Some of my designs just can't seem to be well converted to using DEAO....
I've done well converting many to DEAO... but some just don't...
there's a depth of field which is deeper with the SSAO24 types...
SSAO24r is what I've been using most...
Pretty please smiley lol

[Jesse]

The SSAO must have quite the whole Z buffer of the big rendering, so this is not possible.

Maybe an additional parameter for the DEAO, to scale the length of the ambient rays to calculate, might give a little option to play with...

[Raydianze]

that might be a possible improvement to create more depth to the raytrace DEAO shadowing...
the SEAO24 types have a very nice depth of field to them....

[knighty]

The SSAO must have quite the whole Z buffer of the big rendering, so this is not possible.

I think it would be  possible with a multiresolution approache: in principle the resolution of the depth buffer needed to shade a given point decreases as we get further from that point. Say one want to render a 2ⁿ x 2^m image with 2^n' x 2^m' tiles. He could maintain a hierarchy of increasing resolution buffers covering the current tile and updating lower resolution levels only when needed. Then while pocessing SSAO, if the currently probed point is outside the current tile he uses the lower resolution level and so on until the lowest resolution level is reached (which anyway covers the whole render).
This approache could also be used to overcome the artefacts at the edge of the picture and also for DOF.

A much simpler solution would be to limit SSAO radius to some resonable value (wrt memory requierments).

[Jesse]

Yes, Knighty, but you have to precalculate the Zbuffer, at least, for all tiles, calculating then SSAO and all the rest seperately.
It's just a totally new procedure, much to much work 

[knighty]

No pressure intended, I'm aware that that would be too much work.

It is not necessary to compute all tiles in this approache. Say we have to render a picture 2^16 pixels wide with tiles 2^10 wide, we just need to mantain 6 levels with increasing resolution. These 6 tiles can be processed independently. one just have to scale "DE stop" value (and maybe other ones). In our example, in the lowest resolution level, the "DE stop" value should be scaled by a factor of 2^6 then 2^5 for the next level and so on until we get to the full resolution level (2^0=1 scaling factor).

There must be some surprises in the algorithms' details that would make things more complicated and therefore imply more (much more) work. So again, no pressure.

[Jesse]

Using a large DEstop value for the downscaled layers is a good idea, used atrous wavelets instead to generate those.
Would be a new function, comes to place 6 or so of the more work needing functions on the todo list 

I really feel no presure, so this is of course no problem.  Dunno if i can fullfill many things, have some other issues so i don't start bigger things by now at all.  But this ideas are always welcome, others might can profit from them too.

Thx!

[mondaymorninglunatic]

I thought about this too - how to improve image quality in big renders.

Well - it is possible to render one single image till 7000x7000 resolution. (Or a little higher).

The SSAO-Method is looking much better than the DEAO-Method. I had the idea to calculate the image first with the maximum resolution (~7kx7k) with SSAO and then as a tiled image With DEAO.
Later the SSAO-Image can be upscaled to the big render resolution and used as a second layer over the big-render-image.
In graphics Programms there are usually different methods of layer-blending (multiplicate, min, max, ...).
I guess multiplication or minimum would be a good blending mode.

The combined picture would have all the structure-detail of the big-render-resolution and a little bit of the ambient-light data from the normal data.

I will try experimenting with this idea a little bit, when i am finished with my current big-render-image. But that might take a while still and i wanted to share my idea rightaway.

edit:
i experimented a little with the idea - but only in low res until now. I first rendered 2 images of the same fractal each 1000^2 resolution. then i reduced the SSAO image two 25% (250^2 resolution) and upscaled it again.
with big renders the maximum SSAO image would be about 7000^2.
So i now have a sharp DEAO-image and a blurred SSAO one. I brightened the SSAO and reduced contrast.
I multiplicated the images.



Uploaded with ImageShack.us