GI denoiser reads its history half a texel off (ignores last frame's jitter) #1614
Labels
No labels
amd
avatars
bug
builder
cannot-reproduce
crash
epic
feature
from-viewer
gl-removal
known-limitation
monitoring
parity
perf
platform
priority
critical
priority
high
priority
low
priority
medium
provenance
rendering
rt
status
in-progress
status
testing
status
up-next
triage
ui
upstream
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
sky_line/Slipstream#1614
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
The GI denoiser's temporal pass maps motion vectors to last frame's unjittered position, but the history was traced on last frame's jittered sample lattice over a jittered image. Every history read is up to half a texel off (clean_dn_temporal.comp, the reprojection; GI doesn't set sampleGridHistory). Result: a sub-texel resample every frame (softer GI) and edge taps that pass or fail the geometry tests with the jitter (shimmer), at full resolution too. The stochastic shadow denoiser already corrects this.
From #1613. The default lag fetch (clean_deferred_light.comp deferGiFetchLag) has the same offset and stays a follow-up.
Shipped in 1.6.0 Beta 4.