Hi team,

Is there anyway to control the number of deep samples output by the Karma deep camera output?

I am seeing very large numbers of deep samples generated, even with simple hard surface elements. In this example here are a couple of platonics rendered with Karma deep camera map.

When I sample the deep data in Nuke there are as 60+ samples per pixel. This is a lot of redundant deep data that makes the deep files slow to work with. Arnold for example has various alpha and depth tolerance controls to heterogeneously modulate the number of deep samples per pixel based on various depth and alpha heuristics.

Are there any Karma deep sample rate settings I'm missing?

There's "DCM Compression" parameter on Render Product LOP which is set to 5 by default (higher the number, lossier/smaller the result). If you're enabling deep via karma LOP, you'll have to dive into the LOP and find the product node and change the property there.
Can you sample other areas of the image (say, the grey face right below the one you sampled, or the blue platonic in the background) and see how many samples there are? I suspect the reason why there are so many samples in that particular pixel is because it's overly lit, and discarding more samples introduces too much contrast between the actual data and reconstructed signal which can't be tolerated at the default compression setting.

dlee
There's "DCM Compression" parameter on Render Product LOP which is set to 5 by default (higher the number, lossier/smaller the result). If you're enabling deep via karma LOP, you'll have to dive into the LOP and find the product node and change the property there.
Can you sample other areas of the image (say, the grey face right below the one you sampled, or the blue platonic in the background) and see how many samples there are? I suspect the reason why there are so many samples in that particular pixel is because it's overly lit, and discarding more samples introduces too much contrast between the actual data and reconstructed signal which can't be tolerated at the default compression setting.

What image planes are used for computing the tolerance?

jsmack
What image planes are used for computing the tolerance?

It uses every Render Var channel entered into the DCM product.

how would one generate only deep opacity map, aka DSM?
which rendervar do I point DCM Render Vars parm to?

I've mostly used only DSM in Mantra due to image sizes, so would be great to have easy way to do that in Karma also
that would also avoid oversampling based on light as only opacity values would be considered over depth

tamte
how would one generate only deep opacity map, aka DSM?
which rendervar do I point DCM Render Vars parm to?

I've mostly used only DSM in Mantra due to image sizes, so would be great to have easy way to do that in Karma also
that would also avoid oversampling based on light as only opacity values would be considered over depth

It sounds like deep images need to be converted to point clouds so they can be binned properly with a fuse sop or something.

tamte
how would one generate only deep opacity map, aka DSM?
which rendervar do I point DCM Render Vars parm to?

If you leave render vars field on Render Products LOP blank (or if you're using karma LOP, the DCM Render Vars field), it will only use opacity/Of channel.

perfect, thanks dlee

Hi all,

Thanks for the great discussion. I was able to successfully reduce the number of deep samples by editing the DCM compression setting.

dlee, here is another sample on a less lit area from the original image (DCM compression default of 5) still, quite a high number of samples at 22:



Inside the Karma ROP I was able to add an "edit render product LOP" to set the DCM compression to 10. (FYI the "DCM Render Product lop" inside the "karma render properties lop" appears to be using an old version of the render product lop and doesn't have any karma settings, i.e it's not possible to set DCM compression there, H19.0.622).




This resulted in a much reduced file size, and reduced deep samples:




I will submit an RFE to bubble the DCM compression setting up to the Karma ROP. I agree a deep to points and points to deep SOP would be nice!

Thanks!