A Unity ID allows you to buy and/or subscribe to Unity products and services, shop in the Asset Store and participate
in the Unity community.
Discussion in 'Global Illumination' started by andrei-ichim, Jan 8, 2016.
And on the beta? It's not in b20, right?
No, its in b21.
So... about that patch
Works great in B21. . . . you just make me a very happy artist
next is proper mixed mode please
Still having severe issues here. I rebaked the scene in my current project and things got worse, not better, so I created this test scene to see what was going on:
Static object on the right, dynamic on the left. Two directional lights, one red (intensity 1), one green (intensity 2). Seems all probes now have a spot of light exactly opposite the main light direction, and the ambient lighting is still weird; it's too dark and too blue in this case.
Couldn't get the second image to upload in the last post, but here's what stuff looked like in my actual project's scene. The grass and rocks are dynamic, everything else is baked static:
The lightprobes are quite clearly very different than the lighting on the static objects would suggest they should be.
This makes me happy to see
Trying beta 21. The results are a lot better than before, but the issue still remains in areas were normals are tight, especially in creases and normal maps (colours are opposite of the light/emission source):
It could be better, but the results are at least useable so long as you don't increase light and emission intensity too much. I hope this isn't the final fix. We're so close!
But.. what about 5.3's patch? delayed?
Yes, it got a bit delayed. It should go live in the next 10 mins or so.
5.3.5p3 is live now, https://unity3d.com/unity/qa/patch-releases
It's a lot better now, but I'm afraid there's still an issue left. Lights apply light to the opposite side of objects. These tests are from 5.3.5p3
Here's a more dramatic example, scaled up the lightprobe gizmos to show how they react.
The big sphere on the left is lightmapped, the one on the right is lightprobe lit.
After updating to 5.3.5p3 and re-baking my characters look pretty much the same (slightly burnt on the bottom) My probe gizmos look messed up which isn't encouraging.
And I'm also getting this error after hitting play
"Error adding Enlighten probe set: Failed to allocate probe set 596524d671ced247b3d770c14006ac61"
I think we just expect them to work in a different way than they would do in real life. If you look at the Unite 2016 keynote, you can see that its normal to get weird light wrapping on the opposite side of light sources, look for yourself, you can clearly see what i'm talking about
@monkzoren In that case Unity lightmapping is broken, and needs to be fixed.
Estimating by the forum threads on this forum, I think that is the case yes.
EDIT: Looking forward to the progressive ligthmapper!
Indeed, funny how light is caught directly from the opposite direction of the light source, not from the actual geometry that the light is bouncing from. Oh well, I guess physics might not be the best model for light transport after all.
I have done some test and for me lightprobe now reacts as they should.
No problems here.
Thank you Unity !
It's a ringing problem known from spherical harmonics, It's like they didn't use the code to counter act it, it's a known issue of SH with known solution ...
btw just want to check is everyone still getting messed up probe baked in gamma space?
my probes mostly broken when using gamma space, but normal in linear space.
Probes are finally in the realm of usable, so hurrah!
They still do seem a bit weird though. similar to what @Owers saw. I did a comparison with 4.
Found a weird bug (# 804898) with the new light probes - if you're getting some weird results like this, make sure your light-probe-volume encloses all your lights - seems under some circumstances un-enclosed lights will apply to all probes regardless of range.
I have a fix for this "Error adding Enlighten probe set" error coming to 5.3.5p5. In the meantime, as a workaround, you can clear your lighting data asset and then clear the GI cache.
Thank you for reporting, I'll take a look.
How does it look when you clear the GI cache? I tried your scene and it looks fine, so clearing should help. This is because the patch 3 version in some cases reads stale data from the cache. 5.3.5p5 will invalidate the cached probe files and rebake what is needed.
No change with clearing the GI cache and re-baking (also, this is on v5.4.0b21). I added this to the bug report, as well as some details about how it seems to be related to area lights.
I've attached my test project here, in case anyone else would like to see if it occurs for them...
@KEngelstoft Do you have any updates on the issue of the lit area opposite lightsources?
I got an email that my bug was not reproducible anymore, but when I opened the testcase I submitted and baked in 5.3.5p3 I got the backside lit bug. I wonder if the fix was tested with the testcase supplied, or with other cases.
(The email said it was not reproducible in 5.4, but afaik the fix was backported to 5.3.5p3?).
In testcases from the fix posted here, usually several low level lights have been used, which doesn't make this bug apparent. However, it's not representative for many scenarios on mobile where only lightprobes are used to light dynamic objects, and much more dramatic lighting is used. In any case it should be clearly visible in the testcases I supplied.
Here's a typical use-case for our mobile lighting, a dark dungeon entrance. The problem is very visible even when not using "burnout" level lights.
@PlatformSix I had a similar issue where there was a weird blue/yellow color tint in my light probes.
Thanks for posting a way around the issue. Enclosing all lights with the light probe volume fixed it.
@KEngelstoft case #805202 if you need another test project reproducing the bug.
@KEngelstoft Any new updates? been waiting since January, and the wait/lack of info getting a bit ridiculous
I have this patch installed and tried re-baking lighting and probes, its definitely not fixed.
Here's my result. Half the probes are random primary colors and the other half are black.
I also get probe errors in the console.
I also tried another test scene and Im getting the same problem.
I also cleared bake data. this is Unity 5.3.5p3 Clearly worse than before.
OK!! I just cleared Bake data and GI CACHE and rebaked,
and it seems to be WORKING!!!! OMFG!
[Edit]: Still some issues but its better.
Looks like your Lighting Data Asset is baked with a previous version. Did you try clearing GI Cache and rebaking? The 5.3.5p5 should remove the need to do this clear.
Yeh I cleared the lighting data and GI cache and i got the above result its good now but there is still some issues which have already been pointed out by other people in this thread.
http://docs.unity3d.com/Manual/GICache.html explains how to clear the cache.
Hey guys. I have the same problem as everybody here, so I switched to 5.3.5p3 hoping that it would be fixed there. I cleared all lighting data, cleared the GI Cache as explained above, pressed Build, and, three hours later, I got this error for each Light Probe I have in my scene:
This is it. There's no other info in the error.
Also, the probes (and so the dynamic objects) all look exactly like on the screenshots SteveKouts and Shizola have posted.
I repeat, clearing the GI Cache from the Preferences didn't help at all. Does anybody have a solution? What GI Cache settings do you have? In mine compression is enabled by default, and I'm not sure it should be, but I left it as it was.
@krides Do you have area lights in the scene?
Gi Cache compression should be left enabled, it doesn't have any effect to turn it off, except it will take more disk space.
No, zero area lights. However, I have a whole bunch of emissive materials, and also I'm using both Precomputed Realtime and Baked GI (the latter only for two directional lights).
Hmm, could be a different issue. Would you mind filing a bug report so I can investigate your scene. I haven't seen this before: IsFinite(indirectProbeData[coefficient])
I am currently doing another bake (it will probably take another hour or so). If the issue persists, I'll file a report. Thanks!
UPDATE: Okay, the bake is finished. The light probes are broken like before, but I no longer have the IsFinite error this time.
UPDATE 2: Uploaded a bug report. Case 806217
Cleared the GI cache/ lighting data, some probes are fixed,some remain broken.
I tried the suggestion of 'encompassing all lights' with probes, but it had no effect.
The errors occur based on proximity to the coloured light source ( point light / no shadows / baked / directional specular).
These probes are newly created- post update.
Has anyone looked into environment cubemap to light probe conversion? Might be a more bug free way to get decent looking lighting.
Moving the light somewhere else fixed the issue.
I then moved it back to the original spot, and it went back to being broken.
I finally offset it a tiny bit from the original spot and the results are now correct.
It looks like some part of the cache isn't actually cleared when I clear lightmap data +GI cache, and I have to "jiggle" my lights to fix the issue.
I can rebake lighting and get totally different probe results, nothing seems consistent. Is there any other additional cache or file I should be aware of?
I fix it in one area, it breaks in another- I must be missing a piece of this puzzle.
clear baked data in lighting panel Build dropdown menu then clear gi cache then rebuild lighting
@KEngelstoft Any comment/tips on the light opposite light direction?
Here's an image from a simple testcase.
Create a lightprobe group at 0,0,0
Create a light at 0,0,0
Bake the light wiht Baked Gi
All probes get significant lighting opposite the light direction.
Hi @Futurerobot! The fixed light probes are now matching Unity 4.x behavior. I have recreated your scene and baked it using Unity 4.7.2.
Unity 4.7.2 gamma color space:
Unity 4.7.2 linear color space:
The bright spot is caused by the intense direct lighting causing a ringing artefact on the opposite side. This is an inherent limitation of spherical harmonics. We can't fix this easily without going for a different reconstruction algorithm.
You can try the fancy L1 reconstruction code that I posted earlier in this thread and perhaps you can get rid of most of the ringing but the shading cost is much higher so it might not be suitable for all platforms. Keep in mind that the light probe SH axes is now swapped around so you have to swizzle them in the shader.
We might implement the fancy L1 reconstruction code into Unity at some point, if the demand is there (vote on feedback if you need this).