Hey there

I'm kind of new to the Houdini forum, so please pardon my potential lack of etiquette!

I've been finding that rendering mantra volumes with noise sample refiltering (gaussian -w 2 -r 1) crushes the alpha of the volumes. The dark (less dense) areas seem to disappear completely in the alpha channel.

The issue occurs with raytracing or PBR, but goes away with Micropolygon Rendering.

Additionally, the problem goes away for raytracing, PBR, and Micropolygon Rendering when switching back to gaussian -w 3.


I'm not sure if this is a bug, or something that I should have just known about.

Any thoughts?

Best wishes

Jeff

splintersilk
I've been finding that rendering mantra volumes with noise sample refiltering (gaussian -w 2 -r 1) crushes the alpha of the volumes. The dark (less dense) areas seem to disappear completely in the alpha channel.

Drat. Any chance you could post an example or submit a bug?

Sadly, there were a few (unfortunately rather common) cases that I didn't get around to addressing with the noisy sample refiltering, regarding consistency of deciding which samples to refilter between adjacent pixels, especially if most of the samples are black and a few of the samples are bright, as is often the case with very indirect lighting.

As for alpha, I seem to recall that the refiltering couldn't be applied to it, else there were issues with getting alpha above and below 1 on opaque surfaces, though maybe it was just refiltered separately from r, g, and b. What you described doesn't sound like either of those, though, so it's quite likely a bug. :wink:

The issue occurs with raytracing or PBR, but goes away with Micropolygon Rendering.

Yeah, micropolygon probably has less / different noise to begin with, but doesn't end up with the same final results as raytracing / PBR in general.

Additionally, the problem goes away for raytracing, PBR, and Micropolygon Rendering when switching back to gaussian -w 3.

Yep, the default is off for the refiltering option, and -w 3 is a slightly larger filter width.

Hi there

Thanks so much for the reply, and for your thoughts on it. Good information to have!

I actually packaged a hip file, screen captures and two resulting EXR files that highlight the concern and sent them to support earlier today (letting you know to avoid duplicated effort at your end!). I'll add them to this post.

To quote what I passed on to support:

The ultimate problem with this issue that I'm facing is that with gaussian
-w2 -r1 option I can no-longer unpremult and re-premult in comp and get the
same result.

I've sent two exrs that demonstrate the concern. gaussian 3
unpremults/repremults to the same result, gaussianw2r1 does not.

I hope this info adds a bit more clarity to the situation.

Many thanks and best wishes

Jeff