Unity 5.3.1p2 Light Probe issue: dark spot in the opposite direction of the directional light

@KEngelstoft Thanks for the update! I guess the reason it's a bigger problem for us now than it was in 4.x was that we could bake in gamma where the problem is much less apparent

Bit of a tangent, but will the option to bake in gamma be implemented? Right now mobile can only use gamma-space, while the lightmaps/probes are always linear. So lights become very different for realtime. (And our lighting setup would again match the 4.x look.)

 

Yes but it's somehow worse in 5 as you can see in the example I posted above. Both are in Gamma mode and the Unity 5 probes are much more contrasty.

 

I've found an easy way to reproduce my issue.

I created a new scene, used static unity cube objects to create a room. I added a bunch of probes and a point light. Lighting looked mostly normal. I then add a spot light facing the wall, and my I get a strong black artifact on the probe near my other light source.




I then remove the texture map, specular and gloss contributions from the material and still see the artifact


Increasing the albedo brightness to white removes the artifact


It seems that light probes are receiving weird results when it comes to bounce light.

( I also tested with no crack in the room, and the results are exactly the same )

Because our indoor scenes feature a ton of directional point lights we are receiving this checkerboard artifact on many of our probes, causing a chrome effect on characters.

It seems as though, instead of adding the bounce, it's replacing the light that was there.

 

It seems like Unity 5+ always bakes in linear space, and not gamma space. So the unity 5 gamma bake test is actually linear, which makes the artifact more apparent.

Which is a different issue, especially for mobile where linear is not available in unity: https://issuetracker.unity3d.com/is...ma-slash-linear-lighting-inconsistent-results

 

So it yet again comes back to Gamma baked stuff being out of whack in Unity... Sigh.

In any case, I made a bug report for this specific instance:



Case 806665

 

So, anyone found a fix/workaround of any sort? Patch seemed to mostly fix the issues I was having in one scene, but started work on another tonight and issues are worse than ever. This is a single room with a single light source. Clearing cache, moving light source around, of course clearing lightmap... nothing seems to help. I guess next step is to try just placing the probes again to see if some voodoo magic makes it work? I dunno, but I (and probably everyone in this thread) would greatly appreciate a fix. There's no way I can ship a game like this.

EDIT: Ok, solved the issue. Kind of. If you turn on realtime GI before baking, the color/shadow artifacts seem to go away.

 

Man. . . . i totally missed the shadow/AO accuracy from beast . . . less leak happened etc

 

I think my post is related to this, no feedback from Unity team :/

http://forum.unity3d.com/threads/new-light-probe-bug-in-5-4.412382/

 

Do bug reports (with repro scene) and post the case numbers here.

 

I will see if I can reproduce in a smaller scene

 

Happening to me too

 

I've updated to 5.3.5p7. The issue is still there.




Left groups are static and lightmapped. The right ones use LightProbes.

Clearing GI cache and rebuilding the lighting does not change anything.

 

Could it be possible to see Unity's Probe test scene? cause there have been lots of "Fixes" coming our way, since January. I not tested patch 7 yet myself, but seeing what the last post reported, it does not look promising

 

That's not what was fixed. What was fixed is the bug reported here I believe : http://forum.unity3d.com/threads/un...-directional-light.378516/page-3#post-2680721

 

Man seeing all this enlighten problem in Unity it really frustrating, i wonder is UE4 also encounter the same problem?

 

It looks like the problem is fixed in 5.4.0f3

 

WOHO Im not getting the wierd cross thingy in 5.3.6p1!

 

5.4.0f3 here.... and its still there

 

okay, thank you for the reply...
so, if I understand correctly, to reduce this artifact, i have to reduce the light intensity and then increase bounce boost in the baking, in order not to lose the illumination of the room?

 

More evenly lit environments make the artifact much less pronounced.

More dramatically lit scenes with one strong light lighting up the whole scene (which is something I like to do, sadly) are the worst case scenario for the current implementation of the lightprobes.

I think the decision to make baked gamma space look like linear gamma space was a bad one. First of all it doesn't look like linear, it looks worse it has even less contrast and has a number of other issues. And now it seems it's also the reason why Gamma Unity 4.x probes look so much better than Gamma Unity 5.x probes. *sigh*

Linear space is good, because it produces correct, but somewhat "flat"(/neutral) results. Linear gives you more "range" to work with, so it's better if you are going to be using tonemapping and/or color correction. Introducing more contrast with post processing is much easier than having an already very contrast-y scene and trying to reduce contrast (you'll be trying to recover highlights and shadows that aren't even there or are very crushed). And since Linear is meant for mid/high end platforms where you are bound to be using Tonemapping and/or color correction, everything works and looks great.

But making gamma space, which is mainly used for mobile (where post processing is doable but takes a lot of performance) having baked lighting have even less contrast than linear, means you are more or less stuck with weird looking lighting and no tools to redeem the situation.

I have ended up editing the baked lightmaps in image editors to introduce some contrast to them, which is less than great since if I try to push things more than just a tiny bit, I break coherence with the light probes (not to mention I introduce extra banding), which is again, less than great. Plus it complicates my workflow when I need to rebake my scenes.

I am hoping that the progressive lightmapper will produce drastically different results, but then again, since the current situation seems, up to a point, intentional, maybe I shouldn't have too much hope.

*sigh*

 

I hv this issue.. Without bake maps i hv light in game, after i bake all point lights it all turns black.. Unity 5.4.1f

 

Would this be 2018.3? Says so in the roadmap

 

Will it be backported to LTS?

 

But you could argue its a fix for broken light probes though

 

They could but they won't because then everyone would want new things backported. And this would cause endless bugs and... oh... be just like upgrading to a new version of unity........ :O

 

Oh, we are updating, we always have 3 branches going, production branch (2017.3.x something), Dev branch (2017.4.10) and migration branch (2018.2.6). Both Dev and migration branch are constantly being updated to latest version but right now we have no reason moving to 2018 since we are stuck on old PBR pipeline and currently have no real reason moving to beta version of ECS. But we do have alot of strange artifacts on dynamic items due to light probes

 

Hey, I am still having the original problem with really bad dark/blue spots. Rebuilt light several times, There's only one directional light in the scene set to Direct 1 / Indirect 1, Environment for Bake is also set to 1. What am I doing wrong?

 

Did you tick the tickbox on the light probe group before baking?

 

Sorry but I am sadly not exactly sure what tickbox you are talking about, the light probe group itself has no settings on itself, has it?

EDIT: Ah I am still on 2018.2, you were probably talking about the Beta branch right? I am gonna test that out next

 

Inb4 light probes in unity 4 still look better than they do in 2018+...

Honestly, just have a legacy lightprobe math option or something.

 

It's not that simple.

Also, this has been more or less fixed in 2018.3.

 

Has it? I haven't seen a screenshot of the updated de-ringed results and how they compare to 4.0 results. 5 minutes of Googling for them hasn't been fruitful either. I'd test it myself but our project is locked in on Unity 5.

 

Left is Unity 4 and right is deringed lightprobes in 2018.3. The lighting of the scene is kinda different (the main light source is brighter in Unity 4) so it's hard to make a 1-1 comparison (and I don't have time to actually try to match them better right now).

In short, with deringing, the ringing is completely gone, but they also tend to have a bit less contrast (more pronounced here because of the lighting differences) than Unity 4 probes.

 

Both in gamma or linear space?

 

Both in Gamma.

 

Ah okay, yeah you did mention the light intensity are different, it's just the new one looks like tonemapped. Alright thanks, i'm just curious