Understanding FG cache files

[voxel]

1) I rendered a FG-only camera pass (a set) in Maya 7.0.1 to generate a large FG cache file(aka Rebuild off).

2) I used this FG cache (aka Freeze) in a scene with characters walking/dancing over the set.

Now... I've known that the FG points in the FG cache contain more than just position - lighting/color is store also (not a good thing to be honest).

So when I render (2) - the occlusion effects normally created by a character against the set disappears. I think this is due to the FG cache containing all the FG points ever needed to render and so no new points get generated and no new occlusion/lighting values get computed.

My questions are:

1) Should I reduce the sampling density of the FG cache so that in my final render - new points will be generated? I'm worried this will lead to noise.

2) Is there a way to get rid of the illumination in the FG cache? I want to use it a sampling grid and not a point cloud light map (which it seems to be).

Is my understanding of FG cache files wrong? Any thoughts?

[bart]

FG points are stored in world space in the FG map. Indeed, it is a point cloud like storage; I like to say its like a bucket of points with light information. BTW, maybe I'm misunderstanding, but how would an FG point be useful if it only had position but no light value?

Freeze is designed to be used with camera fly-through like animations. Moving characters means moving illumination. So those occluded FG points are calculated but are averaged with the other ones, likely more of them from frames without the occlusion, if your character moves quickly.

You could still try to freeze areas of points which do not have moving illumination.

Two ways to do that.

1) Make a pass that does not contain FG points (FG receive off) in the moving illumination areas

2) Use fg_copy's bounding box option to extract the FG points that you want to keep, and possbily remerge several of these extractions together with it, as well.

[voxel]

BTW, maybe I'm misunderstanding, but how would an FG point be useful if it only had position but no light value?

I'm almost certain noise due to FG without a FG cache (in an animating scene) is due to the "random" generation of FG sample points in screen space. Frame to frame coherence of FG sample points is non-existent without the FG cache unless you oversample like mad - I do it in Maya 7.0.1 with radii settings of 0.1 and 1.0 (subpixel FG point generation).

I originally thought FG maps were solely used as a sampling grid in world space that is predictable and reusable, but I'm wrong. However, the FG cache does work amazingly well(i.e produces noise-free animations) even with horribly low FG ray settings and discontinuous area lights.

90% of the time, the FG sample points would evaluate to the same illumination value so I can understand the benefits of storing the illumination value. But for 10% of the time and general use, I'd want the FG points for "sampling-only."

I want to use FG maps because they lower render times considerably and also reduce "FG noise" due to frame incoherent FG sample point generation.

(1) sounds like a workable idea, but a little human-time heavy.

As for (2), I'd want fg_copy to do the OPPOSITE and remove fg points from the bounding box I provide in the command line. Given a set + FG cache of it, I could estimate where characters would walk through and provide a gross bounding box for FG points I don't want. Is this possible? Or I do write something to invert my "negative" bbox into a collection of "positive" bboxes?

[bart]

A sampling only FG point cache wouldn't really speed up anything. The time is spent in the material calculations occuring when the primary FG rays hit things.

That said, it is an interesting idea for noise reduction. mental ray does not use random, but rather, qmc sampling methods. So, we're not sure how much reduction could be accomplished.

Very advanced users might be able to set up seeds for the QMC to get identical locations, but tests indicate that there is more to cause of the noise than the location of the FG points. In other words, our tests indicate that it is not the locations of the FG points that cause the noise.

[philipp]

Maybe al little offtopic, but I stumbled over this thought (and I havent spent too much time with it yet, but I will if it seems to be worth the time...):

Would it be possible to "abuse" a FG cache to store samples of glossy reflections (and not indirect illumination) and use the algorithm to interpolate between those? (Like VRays ability to speed up blurred reflections by not calculating them for every sample, but look them up from a cache and interpolate where necessary...) .
Just a thought, maybe someone has experience with it?

[bart]

It might work if you could get the glossy into the FG map.

BTW, you could store glossy reflections in a light map. Then you set the density of control points by the combination of vertex resolution, and lightmap resolution.