When rendering a Z Depth (Pz) pass, there are these rectangular boxes (voxel boundaries?) that show up around the edges of the volume. You can see the result in the attached images.

I have tried setting filters for the deep raster output Pz, and the results did not change much. I'd like to know if these box shapes are expected, and if they will affect typical compositing tasks like blurring along Z for DOF, etc.

If these boxes are going to be a problem, what are the deep raster output settings that can smooth them out?

EDIT
p.s.: The Pz pass has been color-corrected down so the Z gradient is visible.

I guess they are volume voxel boundaries. As far as I can work out it is difficult to determine the depth of a volume because of its density. At what point does the volume density become high enough to consider it a definitive z distance? If you integrate it over the total marching distance then it makes no sense (?) and if you take the first ray “hit” then you get the voxels like you see in the image.

You could conceivably set a low density threshold below which a ray hit is not considered valid to count as a distance. That way you could presumably soften the blocky look.

It's an interesting problem though and perhaps there is a answer to it. Maybe there is a way to combine density with distance but then you'd get at least a 2 component variable (?). I can't immediately see how it could be done in a one dimensional way.

But then, I don't really know enough about it to know for sure.

Soothsayer
I guess they are volume voxel boundaries. As far as I can work out it is difficult to determine the depth of a volume because of its density. At what point does the volume density become high enough to consider it a definitive z distance? If you integrate it over the total marching distance then it makes no sense (?) and if you take the first ray “hit” then you get the voxels like you see in the image.

It's an interesting problem though and perhaps there is a answer to it. Maybe there is a way to combine density with distance but then you'd get at least a 2 component variable. I can't immediately see how it could be done in a one dimensional way.

But then, I don't really know enough about it to know for sure.

Macha, it's great to hear from you!
You are definitely thinking about this in a different dimension than I am. No pun intended.

I'll be honest and say I was only trying to get a Z image for comp., and wasn't really thinking about the geometric/spatial implications of recording voxel density in an image.

This reminds me, wouldn't the deep camera map technique be useful here?
If I understand the problem you presented correctly – i.e. where a micro-voxel is not a sample point in space, but a volume with depth; so recording a single Z value per pixel does not represent the micro-voxel properly – woudn't deep camera map allow an accurate Z view (multiple Z values) into the volume?

I'm trying to render this with the Deep Resolver parameter in the Mantra (11) rop now. Will post back results.

IndyZoneCo
This reminds me, wouldn't the deep camera map technique be useful here?

Yeah Indizone, I'm still here, hehe.

I haven't used or heard much about deep camera but it sounds a bit like what I refered to when I said “integrate over the total marching distance”, ie sum up all the distances.

It sounds like a possible solution, but I know too little about deep camera maps.