Fix a flickering issue related to moving shadow receivers (case 1302392) and a refactoring to avoid computing the same value for every effect. #3039
Conversation
There was a problem hiding this comment.
Tested multiple settings for the 3 raytraced effects concerned (RTGI, RT Shadows, RTAO) in the repro project and in another bigger real world project.
The ghosting seems to have completely disappeared when the "receiver" is moving in RTGI / RTAO and RT Shadows except RT Area Light shadow. It's the only case that does not seem affected by this fix. (see gif)
Despite ghosting being fixed, I have the feeling it's maybe a bit to "agressive" on discarding the history, is this possible to fine tune the threshold for history validity ?
I've noticed that the history gets discarded instantly if the receiver is moving, even if we cannot see it moving with out eyes. I think it would be a bit better to increase a bit the tolerance to be able to NOT discard history if it's moving very slowly, some ghosting can be tolerated especially if it's moving really really slowly.
I tried fiddling around with the constant in TemporalFilter compute shader but not luck
You can see an example of what I'm saying HERE
Difference BEFORE & AFTER
Before
After
|
No, not possible to "tune" it, we don't have access to that piece of information |
There was a problem hiding this comment.
Concerns about unaffected area lights won't be fixed in this PR, it will be the object of future work.
As for the very low moving threshold, it can introduce some new issues, (unwanted/unjustified noise if the receiver has a very low velocity basically) but I believe this is still an improvement compared to previous implementation.
Approving ✔️
…in world space (case 1302392).
…or the whole frame and not per effect.
anisunity
force-pushed
the
HDRP/rt-shadow-flicker-history-validation
branch
from
d1c4f8b to
0cb8e6a
Compare
sebastienlagarde
deleted the
HDRP/rt-shadow-flicker-history-validation
branch
|
Unfortuantely can't be a trade off. Either it ghosts or it doesn't |
…92) and a refactoring to avoid computing the same value for every effect. #3039
* Fix issue with register spilling in light list shaders #3016 * Fix a flickering issue related to moving shadow receivers (case 1302392) and a refactoring to avoid computing the same value for every effect. #3039 * Merge Hd/bugfix #3047 * [HDRP]Improvements and fixes on Volumes Components Inspectors #3072 * Fix XR depth copy and MSAA #3075 * [HDRP][Compositor] Fix issues with compositor's undo #3100 * Hdrp/docs/fix 1305538 (#3112) * Added information about using recursive rendering and other effects * Fixed formatting * [HDRP] Merge Hd/bugfix #3120 * Formatting Co-authored-by: FrancescoC-unity <43168857+FrancescoC-unity@users.noreply.github.com> Co-authored-by: anisunity <42026998+anisunity@users.noreply.github.com> Co-authored-by: alex-vazquez <76204843+alex-vazquez@users.noreply.github.com> Co-authored-by: Fabien Houlmann <44069206+fabien-unity@users.noreply.github.com> Co-authored-by: Pavlos Mavridis <pavlos.mavridis@unity3d.com> Co-authored-by: Lewis Jordan <lewis.jordan@hotmail.co.uk>
…92) and a refactoring to avoid computing the same value for every effect. #3039
* Fix a flickering issue related to moving shadow receivers (case 1302392) and a refactoring to avoid computing the same value for every effect. #3039 * The RTAO's history is now discarded if the occlusion caster was moving (case 1303418). #3343 * Update CHANGELOG.md * Add new API to CachedShadowManager and fix WouldFitInAtlas #3387 * Added master stack documentation for HDRP (#3388) * Added landing page and context/blocks overview page * Added vertex context page and snippets * Added the fragment context page and supporting snippets * Added the decal master stack and the contexts intro * Updated toc * Added vertex context doc * Update TableOfContents.md * Added unlit doc * Added lit master stack doc * Added eye master stack doc * Added fabric master stack * added hair master stack * Added stacklit master stack * Added snippets to support shader stack work * Fixed some formatting bugs * Removed old links * Updated links * formatting Co-authored-by: Lewis Jordan <lewisjordan@unity3d.com> Co-authored-by: sebastienlagarde <sebastien@unity3d.com> * Fix Sub-shadow rendering with cached shadows + Fix PCSS filtering issues on first frame or cached shadows #3396 * Reset ambient probe upon switching to very different skies bis #3423 * HDRP Graphic Test Fixes for Metal (#3408) * Fixed the metal pipeline creation error message and updated several gfx test images. * Rearranged LightData definition to workaround area lights not rendering issue on Intel GPUs on Catalina * Updated 2501 reference image for Metal with the area light fix * Updated 1550 and 2010 on Metal as previous were incorrect due to area light issue Co-authored-by: Jeannette Yu <jeannette.yu@unity3d.com> * Custom Pass error message fix (#3424) Typo in DrawRenderersCustomPass error message. Fixed a few other typos in this section too :D * Add Unity version to what's new 11 (#3432) * Update whats-new-11.md * Updated the table of contents * Shader abstraction cleanup #3443 * Some leftover caused by master merge (#3456) * Fix multiple any hit occuring on transparent objects (case 1294927). #3471 * Colored Shadows update (#3486) * Colored Shadows update Included new information in the Color Shadow property description to tell the user that the Material's Refraction Model has to be set to Thin, Box or Sphere for it to work. * Fixed typo Removed the "the" :D * Revert "Fix multiple any hit occuring on transparent objects (case 1294927). #3471" This reverts commit 0587177. * Update 1219_Lit_Light_on_Tesselation.png Co-authored-by: anisunity <42026998+anisunity@users.noreply.github.com> Co-authored-by: FrancescoC-unity <43168857+FrancescoC-unity@users.noreply.github.com> Co-authored-by: Lewis Jordan <lewis.jordan@hotmail.co.uk> Co-authored-by: Lewis Jordan <lewisjordan@unity3d.com> Co-authored-by: Brandon Fogerty <34072588+brandon-unity3d@users.noreply.github.com> Co-authored-by: Jeannette Yu <jeannette.yu@unity3d.com> Co-authored-by: Vic Cooper <63712500+Vic-Cooper@users.noreply.github.com> Co-authored-by: robinb-u3d <robinb-u3d@users.noreply.github.com>
This PR includes twp seperate changes:
Testing status
I ran the DXR tests locally they are all green.
Opened couple scenes and made sure the reprojection looks fine.(Shadows, AO, RTGI)
Unfortunately, the tests do not capture temporal denoising this would require a bit more manual testing from @remi-chapelain to make sure it behaves normally after the PR.