Render Target Disocclusion Artifacts

New disocclusion artifacts introduced in 5.6 occur at the edges of opaque objects in front of volumetric clouds and fog. See video for a comparison between 5.5.4 and 5.6.1.

Setting r.VolumetricRenderTarget.Mode=1 improves the results dramatically but still leaves some visible artifacts and costs dramatically more due to the higher internal resolution.

Various r.VolumetricRenderTarget.UpsamplingMode settings aside from the default only make the issue worse.

I can see Sebastien Hillaire has made the vast majority of commits to the volumetricrendertarget.usf shader file and so is likely the best candidate for a response.

[Attachment Removed]

Steps to Reproduce
Basic scene with default volumetric cloud material w/ StormClouds parameter set to 0.6 (I have confirmed this issue with an entirely different material but this is the easiest way to reproduce). Add primitive objects to use as disocclusion objects. Track camera back and forth across the objects with the clouds in the background.

Result - visible pixelated artifacts.

[Attachment Removed]

Hello,

I wasn’t able to reproduce this in latest, but there was a commit in 5.7 to fix some disocclusion artifacts that may be what fixes it:

CL#44160343 Volumetric Cloud - Fixed bad disoclusion happening for animated object over background.

If that doesn’t fix the issue, can you please provide a simple repro in a template project that we can use?

[Attachment Removed]