How view dependent is Light Cache?

[potsked]

I am working on a 2-min plus animation and need some clarification on how Light Cache operates in fly through mode. Here's my 2 guesses as to how it works, please tell me where I'm wrong if you can.

 

Guess 1: it operates identically to IR-Map in "incremental add to map" mode, in that you only need to make sure you have all the surfaces covered adequately and any camera view you need can be accommodated by the "map" you build up.

 

Guess 2: All the samples are split up evenly among the frames. Therefore you multiply the typical number of samples a shot requires by the number of frames.

For instance if a shot usually requires 640,000 samples (or 800 subdivs) and you need 100 frames then you need 64 million samples (or 8000 subdivs). This, im assuming, is driven by the LC's view-dependence, in that only the samples taken from a given view can be used for that given view.

 

This problem is exacerbated by my animation which keeps a very tight shot on a nearby object with a detailed background fairly far away. Please let me know if you have experiences with this issue.

 

Thanks

[Thablanch]

Never really did the math...

 

I would not think it is distributed evenly, as in the lc process, some frames are calculation faster than others.. So I would go with guess 1, as it keeps in file info that does not need to be recalculated all the time..

 

http://www.spot3d.com/vray/help/150SP1/tutorials_imap2.htm

[dlparisi]

In practice it operates similar to your guess 1 but I think it's doing something different than the IM "add to incremental". Determining the correct number of samples is a bit of an art, you'll know when it's too low as the GI may start to look sloppy and if it's too high it'll take too long to calc (whatever "too long" is to you). I usually start with triple the number of samples for a 1 or two room animation and do some test renders using just the light cache to gauge the quality of the solution.

 

For your scene, you may want to try world scale for the LC so the further items get adequate samples, see which is better.

[potsked]

For your scene, you may want to try world scale for the LC so the further items get adequate samples, see which is better.

 

On that very point I still am confused. Every tutorial out there insists you use "World" scale since not doing so creates flicker. But the flicker comes from the changing number of samples a surface gets as your camera gets closer/farther away. If the LC samples were additive, then the camera would take into account every sample taken of a surface and give the same result on any given frame/camera angle.

 

The insistence on "World" scale would seem to stem from sample's dependence on time/angle. I'm gonna run a few tests and see what happens.

[dlparisi]

I don't think that setting it to 'world' avoids flicker, it just creates a more uniform LC. There should not be flicker in either case.

 

Just to clarify, you are running the LC calc only once and not for every frame, right? The flythrough mode runs the LC calc for the length of the animation and should not be run at every frame. For example, calc the LC with the time set to 0-100 frames and set it to autosave then run the Irr Map in Incremental Add (if that's what you're using) for the same frames (maybe by every 10th frame though) using the saved LC and save the resulting IM. When rendering you can turn off the LC completely (unless you have "Use LC for glossies" checked-see the help) since it's stored in the irradiance map at that point. Unless you have animated objects or lights in the scene (you didn't mention your scene) there should be NO flickering due to GI. If you have animated objects or lights you can not use the LC flythrough mode.