Page MenuHome

Possible errors introduced in the new SSS implementation (in the days after it was committed)
Closed, ResolvedPublic

Description

System Information
Win10 64 bit, an ATI 7xxx GPU, 24 gigs RAM, third gen i7 processor

Blender Version
Broken: hash 578f70f
Worked: hash da7ddb6

Short description of error
Very recent tweaks to the new SSS model seems to have introduces what appears to be pronounced steps in the terminator when the radius has different values for the red, green, and blue channels. This did not exist in the initial implementation and is far harsher than simple cubic scattering. This being with CPU rendering by the way).

Exact steps for others to reproduce the error
Download the .blend file and render, here's an image of what to expect depending on the builds (I will also note that there were no settings whatsoever that were changed).

.blend

I know the implementation still has a WIP component to it, but I do think the recent tweaks may have made it less correct instead of more.

Event Timeline

Adam Friesen (ace_dragon) updated the task description. (Show Details)
Adam Friesen (ace_dragon) raised the priority of this task from to Needs Triage by Developer.
Adam Friesen (ace_dragon) set Type to Bug.

Here's a better illustration of the terminator error (with a lot more samples together with a curve adjustment to show it is not a product of undersampling)

You can practically see lines in the falloff, I don't think that's supposed to happen.

Sergey Sharybin (sergey) triaged this task as Normal priority.

Please refer to commit notes for rB69dc0c3. The original patch for this report wasn't correct and Burley is not expected to be that much smoother actually for the same mean free path distance.