Feedback Physically Based Sky

Well it is "functional enough" for static environments and high sun/moon altitudes only, but as soon as the directional light is close to or below the horizon it fails miserably. Dynamic day/night cycles are barely achievable due to the visual artifacts and it's huge performance overhead when updating the sky.

There hasn't been any update to the PBS for like 2 years, that's why I started to work on it myself.
I was just sick of waiting for it to improve. (well that never really happend as you can see)

 

U

I absolutely agree with you.
Low light external scenes have often been a Dodge topic, or different approaches combined.
Mostly because of the wish for dynamic ToDs.

Even the PBS doc states that it was only used in the day time scene.

I certainly appreciate your time and studies over the years that I've seen.

With many really wishing to hone in on realtime features and larger landscapes, you certainly took action in this quickly.

 

They actually started to fix some issues, like the flipped UV of the light cookie;
https://github.com/Unity-Technologies/Graphics/pull/6505

I added the fix to my version aswell, but I also fixed the flipped UV's of the planet and updated the sample textures accordingly;
https://gitlab.com/bad-bug-busters/...mmit/b23abba9043f905b918cf541666aeb4485a2db46

EDIT
Related bug report;
Case 1394827

 

Maybe, if you have time, you can send a pull with some of your improvements/fixes. I believe they accept public pull sometimes, if they aren't very big changes.

That light UV was fixed due to a bug report, it was sent ages ago though, if i remember correctly.

 

I allways wondered why there's no visible multiple scattering happening in the atmosphere, although it's being processed in the renderer ("Number Of Bounces" set to 2 or more).

It turned out that the multiple scattering contribution in the atmosphere is zero when the sampling point is in the planets shadow.
It simply doesn't correctly account for the atmosphere depth and treats it as if it's infinity thin.
That's why the sky turns pitch black when it's in the plantes shadow, so there is absolutely no multiple scattering happening where it matters most.

Here's a picture to illustrate the issue.
This time I ONLY rendered the multiple scattering contribution of the atmosphere and the planet itself (which is being correctly lit by multiple scattering btw) for better visualisation.

So it somehow illuminates the planet correctly, but not the atmosphere itself.

This clearly has to be a bug.
Precalculation of multiple scattering happens here;
https://github.com/Unity-Technologi...nScatteredRadiancePrecomputation.compute#L339
I haven't found a way to correct it yet, but I'll try my best to figure it out.
No Idea if I should just do it myself or spend waiting (probably)months for a potential fix from Unity.

EDIT
Related bug report;
Case 1394838

 

Since the HDRP change from version 12.1.0 until version 12.1.3 the PBS is messing with the light if it's the light is turned off or with 0 of intensity it gives this error:

NullReferenceException: Object reference not set to an instance of an object

UnityEngine.Rendering.HighDefinition.PhysicallyBasedSkyRenderer.RenderSky (UnityEngine.Rendering.HighDefinition.BuiltinSkyParameters builtinParams, System.Boolean renderForCubemap, System.Boolean renderSunDisk) (at Library/PackageCache/com.unity.render-pipelines.high-definition@12.1.3/Runtime/Sky/PhysicallyBasedSky/PhysicallyBasedSkyRenderer.cs:505)

UnityEngine.Rendering.HighDefinition.SkyManager.RenderSkyToCubemap (UnityEngine.Rendering.HighDefinition.SkyUpdateContext skyContext) (at Library/PackageCache/com.unity.render-pipelines.high-definition@12.1.3/Runtime/Sky/SkyManager.cs:596)

UnityEngine.Rendering.HighDefinition.SkyManager.UpdateEnvironment (UnityEngine.Rendering.HighDefinition.HDCamera hdCamera, UnityEngine.Rendering.ScriptableRenderContext renderContext, UnityEngine.Rendering.HighDefinition.SkyUpdateContext skyContext, UnityEngine.Light sunLight, System.Boolean updateRequired, System.Boolean updateAmbientProbe, System.Boolean staticSky, UnityEngine.Rendering.HighDefinition.SkyAmbientMode ambientMode, UnityEngine.Rendering.CommandBuffer cmd) (at Library/PackageCache/com.unity.render-pipelines.high-definition@12.1.3/Runtime/Sky/SkyManager.cs:885)

UnityEngine.Rendering.HighDefinition.SkyManager.UpdateEnvironment (UnityEngine.Rendering.HighDefinition.HDCamera hdCamera, UnityEngine.Rendering.ScriptableRenderContext renderContext, UnityEngine.Light sunLight, UnityEngine.Rendering.CommandBuffer cmd) (at Library/PackageCache/com.unity.render-pipelines.high-definition@12.1.3/Runtime/Sky/SkyManager.cs:948)

UnityEngine.Rendering.HighDefinition.HDRenderPipeline.UpdateSkyEnvironment (UnityEngine.Rendering.HighDefinition.HDCamera hdCamera, UnityEngine.Rendering.ScriptableRenderContext renderContext, UnityEngine.Rendering.CommandBuffer cmd) (at Library/PackageCache/com.unity.render-pipelines.high-definition@12.1.3/Runtime/RenderPipeline/HDRenderPipeline.cs:2564)

UnityEngine.Rendering.HighDefinition.HDRenderPipeline.ExecuteRenderRequest (UnityEngine.Rendering.HighDefinition.HDRenderPipeline+RenderRequest renderRequest, UnityEngine.Rendering.ScriptableRenderContext renderContext, UnityEngine.Rendering.CommandBuffer cmd, UnityEngine.Rendering.HighDefinition.AOVRequestData aovRequest) (at Library/PackageCache/com.unity.render-pipelines.high-definition@12.1.3/Runtime/RenderPipeline/HDRenderPipeline.cs:2062)

UnityEngine.Rendering.HighDefinition.HDRenderPipeline.Render (UnityEngine.Rendering.ScriptableRenderContext renderContext, System.Collections.Generic.List`1[T] cameras) (at Library/PackageCache/com.unity.render-pipelines.high-definition@12.1.3/Runtime/RenderPipeline/HDRenderPipeline.cs:1855)

UnityEngine.Rendering.RenderPipeline.InternalRender (UnityEngine.Rendering.ScriptableRenderContext context, System.Collections.Generic.List`1[T] cameras) (at <15e405c9beda435fb78db5a973118bc4>:0)

UnityEngine.Rendering.RenderPipelineManager.DoRenderLoop_Internal (UnityEngine.Rendering.RenderPipelineAsset pipe, System.IntPtr loopPtr, System.Collections.Generic.List`1[T] renderRequests, Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle safety) (at <15e405c9beda435fb78db5a973118bc4>:0)

UnityEngine.GUIUtility:ProcessEvent(Int32, IntPtr, Boolean&)

 

I updated my version to HDRP 12.1.3 and fixed some issues with the volumetric clouds in combination with the PBR fog.
https://gitlab.com/bad-bug-busters/...mmit/b987ba66e1c7ee03ee6455c89b792cf12d3e87d6

The clouds really benefit from the atmospheric pbr fog.



EDIT
Regarding my previous post about the multiple scattering issue;
I still haven't found out what's wrong with it, since it's calculation seems to be correct from what I can tell, but the result isn't.
Maybe it's a precision issue with the LUT's, I don't know.

However I posted a bug report a while ago, still waiting for a response though.
https://fogbugz.unity3d.com/default.asp?1394838_1o4vs80t21e8rtre

 

I reworked the spherical PBR Fog to be faster and less prone to artifacts.
The new one basically uses the same technique as the regular fog and get's it color from a blurred cubemap of the sky itself, this is not as accurate as it used to be, but it's way faster.
It also get's rid of the ugly banding artifacts around the horizon.
https://gitlab.com/bad-bug-busters/...mmit/03bee5f5bb26a8922b5f166fdb89b7eb3a356dd0

No PBR Fog at all, just the regular height fog.


Old method, has a proper scattering result but doesn't blend well with the regular fog and causes banding at the horizon.


New method, approximates the scattering based on the blurred sky cubemap, properly blends with the regular fog and has no banding artifacts at all.

 

@SebLazyWizard out of curiosity did you manage to fix the performance issue when updating the sky setting at runtime?

 

To implement this in my own project, do I just replace the existing HDRP file with the one you linked to on gitlab?

 

Not really, it triggers it's precomputations almost every time you change some of it's settings.
The only thing you can do frequently (not realtime) is to change the position and intensity of the directional lights, which doesn't require additional precomputations, but still comes with a bit of overhead, so only update the lights as rarely as possible (like once every few seconds).
It's documentation got updated recently and provides more info about the precomputations;
https://github.com/Unity-Technologi...ically-Based-Sky.md#warmup-performance-impact

The PBSky is based on an ancient implementation from 2008 which has many drawbacks like the poor realtime performance and it's many visual artifacts.
To be honest it needs a complete rewrite, or atleast an update to later versions of the original creator (Bruneton), which in the end is also a big amount of work.

Yea, make HDRP a local custom package and replace the specific files with the ones on gitlab, that's it.
However if you're on a different version than HDRP 12.1.4 then you have to port the changes over to your version yourself.
I'm not going to support multiple HDRP versions out of the box, it's just my little personal project and not worth the effort.

 

@SebLazyWizard Do you know if this overload also occurs if we change the Vector3 of the "Space Rotation" parameter in real time?
It is not clear to me reading the documentation

 

Atleast the sky cubemap gets rebaked (for lighting/reflections), but that happens every time the sky gets changed anyway.
So again, to minimize overhead, try to change lights/sky settings as rarely as possible.

 

I'm pretty sure you can also set the Update Mode to Realtime with an Update Period of your choosing. This will update the physically based sky based on your interval, leaving you free to rotate your sun and space every frame.

See the docs about Update Mode and Update Period

 

Yes yes... I already use it. I found that with a 0.5 sec update period it has practically no performance impact for me...

 

Does that spread the task over multiple frames or does it spike when it updates?

 

it does all the work in one frame, so it'll spike. I wonder if that will be noticeable if done every half a second though, when playing with high FPS.

 

It'll be noticeable if it puts it below 60 and vsync locks the framerate to 30. Will definitely cause some microstutter.

 

Red dead Redemption 2 used the same algorithm for their sky.... I wonder how they optimized it.... Maybe it's completely prebaked??!

 

Which version of bruneton sky they use 2008 or 2017? afaik the current version of HDRP Sky are using the old 2008 version.

 

I wonder what improvements we could get with 2017: I don’t think it would have a significant performance improvement (and truth be told I haven’t experienced any performance issues with pb sky, I run a 3060 RTX).

You ‘could’ try closing the Profiler window if it’s open, though it probably won’t help much.

Still, I would like to know
@SebLazyWizard why does it need a rewrite, what issues do you see?
@pierred_unity in your opinion, what improvements would Bruneton 2017 bring to Unity?

 

Banding artifacts everywhere, especially at the horizon. Bad visuals at twilight due to missing multiple scattering (well it's there but wrong). This thread is full of examples of it's many issues.

That's the 2017 version: https://ebruneton.github.io/precomputed_atmospheric_scattering/
Web demo: https://ebruneton.github.io/precomputed_atmospheric_scattering/demo.html

There's even a newer and better version out there which is currently implemented into the Unreal engine.
https://github.com/sebh/UnrealEngineSkyAtmosphere

 

So it seems lot of people are agree that bruneton sky are heavy when updating the sky parameter. . . .

 

Hey @pierred_unity !!!
Do you have a road time for these improvements that you are going to make?
What improvements are you talking about? do you have something planned?

I do not ask for a date for the improvements, but will they be in the short term or long term? months or years?

I don't want what I'm going to say to bother you, but we've been waiting for several years now with almost no improvement.

Regards

 

I tried out UE5 and it sky and atmospheric system is silky, buttery smooth and consistent and no artifacts while the PBS sky keeps stuttering the whole game and is useless for day and night cycle... i hope that unity improves it soon.... +1 vote from my side

 

That's odd. It works fine for me. I'm just rotating the sun and there's no artifacts to be seen.

 

Are you by chance changing any of the PBS component settings during runtime? if so, that is why.

instead, you can change your directional lights ( sun/moon) and your exposure overrides on your local or global volumes instead if needbe.

 

Artifact's occur when u move up at an altitude of somewhere 5000m to 8000m during the sunset like this

There are some more... I will provide those screenshots later this week

I am rotating the directional light (very fast) but it lags a lot.... maybe I am doing something wrong?!!

 

Seems fine here, until I move into space at which point it breaks.

 

Fast would be my particular guess there. Directional light movement should be fine in editor of you're moving it fast, and updating probes/shadows at realtime too etcetc during run time you'd probably struggle.

Have you got a video of the lagging that you're able to share?
Certainly peaked my curiosity.

 

Fixed the stuttering by setting physically based sky update mode to every frame(by default its set to on changed)... here are the images of artifacts


about the second image i don't know if this occurs in real life but seems very odd to me... its also very bright at certain angles which makes it even more odd
(all the images are taken when the sun moves below the horizon in under a range of 10 degrees)

 

isn't that reduce the fps a lot? like the sky are updating all the time

 

I guess he meant "Realtime" but with an update period set to anything greater than 0... Doing it every frame is utterly slow.

 

Hmm does change anything for me... performance is same... The advantage of using realtime i see that we get update mode which does help in optimizing it

 

ah right, setting it to realtime but custom update period could help improving the performance

 

From my experience, setting the PBS update to 0.3~0.4 usually gives good performance.
On the other hand, the fact that it looks so bright in the moments after the sun goes down from the horizon, you have to manage by lowering the lux intensity of your directionallight depending on the angle it has. With that the problem of it being so bright disappears.

 

Some of my reported issues finally made it into the issue tracker.
It would be nice if you guys could vote on them, so that Unity might work on them a bit sooner, hopefully.

Incorrect multiple scattering;
https://issuetracker.unity3d.com/is...nt-correctly-account-for-the-atmosphere-depth

Flipped planet ground texture;
https://issuetracker.unity3d.com/is...pped-when-used-on-physically-based-sky-planet

 

Voted both

 

Can someone from unity please comment or give us any update of the PBSky status? will it be fixed, improved, updated, or maybe changed to other skymodel. The current sky have a lot of issue on visual and performance

 

Has anyone found a solution to have both Sun and moon directional lights in scene without breaking volumetric clouds? No matter what setting I use, the cloud always get "lit" by the moon directional light resulting in black shadeless clouds.

 

Volumetric clouds currently only supports 1 active directional light, doubt there's a workaround without changing volumetric cloud code

 

That's likely the case, them would need a check like "if 2 dir lights are present, use the one with higher intensity". So for now the solution seems to add the moon to the space texture or having a textured plane and just switch the directional lights on/off the "usual way" (which feels like that renders the new system somewhat limited and not that usefull)

 

Just for proper continuity and reference.

The clouds came 2 years after the discussion happened.
And about 3 years since first mentioned.

Simply unacceptable for any project.
The studio would have shut down if they depended on it.
(The same happens in other, more crucial features.)

As such my estimate was correct.
Key features that are standard in every other real time engine without any arguments in the forums, are not considered a priority. The solution delivered is adequate, but nothing to write home about considering how many times we heard the "physically accurate" mantra.

 

Well... I guess it is yet another half-baked abandoned unity feature..?

 

Currently you can reduce the number of bounces and update it in every 2-3 seconds to greatly improve the performance

 

I don't know how the atmosphere was released with such a bug



Was it tested at all?

 

It's caused by a "hack" to account for the size of the sun disk, as I explained here; https://forum.unity.com/threads/physically-based-sky.765962/page-3#post-7733619

You can use my modded version of the PBS in which I removed the hack and therefore fixed the visual glitch you're experiencing.
https://gitlab.com/bad-bug-busters/unity/improved-pbs

 

Thank you. I've removed the hack and it is much better right now. I also think, they disabled atmoshperic scattering because of this hack. It destroyed atmospheric scattering lightning on low sun angles.

I don't know why this hack is needed. Bruneton Atmospheric Scattering, for example, has a proper implementation of such sun disk irradiance without hacks.

 

I see one more bug in the atmosphere



Does somebody encounter it? Do you have any ideas where to look, to fix it?