Unwanted highlight with lumen

In a demo scene with 1 directional light and a basic cornell box, I have a weird screen-space effect with lumen as shown here:

The highlight is a bit suspicious. There is indeed a big specular reflection coming from the sun/ground, but I am not sure it should propagates to the nearby green wall.

If I move to a different point of view to hide the specular reflection from the ground, the highlight progressively fades out and disappear:

I checked on the Path Tracer, which I suppose produce “real” results, and the hightlight is not present, regardless of the view point (compare the following Path Tracer image, with the first Lumen image):

I double checked: it comes from Lumen. If I disable Lumen, then there is no highlight.

So, why there is such effect with Lumen ? It looks like to be a screen space effect: just by moving the head, we can observe the highlight to disappear proportionally to the amount of specular reflection there is visible in the scene. It looks a bit weird.

Additional screenshots to demonstrate that this is definitively not “normal”:

Why such glowing area on the green wall ?

Also:

This is a bit puzzling when moving the camera.

There is only the sun in this scene. Any ideas on how to mitigate on this issue ?

I’ve tried enabling Lumen Raytracing with Hitlighting instead of surface cache, but this does not change. Apart from that I am not sure what paremeter would be relevant to tweak to achieve a more realistic result

I’m not an expert but from what I’ve read here on the forum it seems the meshes/objects also need to be in a certain way for a correct lighting. Usually such defects can be seen in certain Lumen Views. Maybe you can post some Lumen views screenshots so we can see if there are any areas in pink colors.

Sure. Here is my lumen scene overview:

No nanite model is involved in this scene. I can’t see anything particularly wrong.

But you gave me an idea by tweaking with the view modes. I decided to isolate the minimum set of lighting features enabled that triggers the effect. Here it is:

This is the “Scene Traces” features that screen to cause the issue.

If I disable this feature, and enable all other Lumen features in this menu, I get a more correct result:

I have to look a bit more deeply about what are these screen traces, and how to tweak them to provide a better realistic result

It’s probably as you noticed in your first post. It has something to do with screenspace. In the past I also got a similar result and researching here on the forum I found out that this screen-space effect should not manifest itself in such a way. I don’t remember now how I fixed it. I’m curious to see what other more knowledgeable people will say.

Ok, I got it.

I do have this workaround for now with disabling Screen Traces (r.Lumen.ScreenProbeGather.ScreenTraces 0). Apparently, based on what the doc says, screen traces are a screenspace pass enabled to optimize things a bit and produce better results for some cases (to solve mismatches between triangle scene and lumen scene), at the cost of control on how the indirect lighting behaves. Which can lead to unwanted results.

But here, I think it is more a “bug”, as the side effect is somewhat obvious for such simple scene. Or maybe this is a corner case, and in practice you don’t have only one directional light with so much high contrast between bright/dark areas (note in the scene, there is NO ambient skylight. This is on purpose, to have such high contrast).

I am interested too to hear about what people will say about how they solved such problem.

1 Like

I’m not sure, but I think I just deleted the offending object - or in your case even more objects perhaps - and recreated them with Unreal Solids but without using scaling. Or maybe using scaling in less extreme ways. And the screen space errors were no longer there. I presumed the previous object had something “wrong” in how the mesh was structured. But in my case the object was pink in some debug views, while it seems to not be the case in your scene.

If I’m rendering something out via MRQ and I’m using HWRT (as the cornell box appears to be), I just disable screen-traces entirely. They’re pretty approximate, and any scene with multiple bounces of specular reflection will show just how poor they are. That isn’t surprising given how the depth buffer is as approximate a scene representation like anything else, but it can make very bad guesses on how to propagate lighting sometimes.

I guess you are right, I did not found a way to improve things, except disabling screen traces. I find Lumen to be really good for diffuse GI, but the reflection/specular side seems to needs to be enhanced a bit

That does seem to be the general conclusion, yes. By and large, the successful UE5 game projects have definitely played to the renderer’s strengths, which aren’t exactly evenly distributed. Diffuse GI is amazing, which is borne out even way back in Land Of nanite, and it was true for valley of the ancients and everything else afterwards.

The Matrix demo was an interesting trick because even though it depended on high-quality specular lighting, every major surface was an almost perfect mirror, which it has no issues with. furthermore, it was a pretty overcast scene, which meant it didn’t need genuine reflection with color bleed as much as it needed specular occlusion, to just block sky lighting in the right places.

We don’t really have any epic project that’s using glossy reflections in any big way, because UE5 can’t do glossy reflections well.