[ki_cz]

Thanks a lot for having a look, hopefully somebody will give us some insight

It's a strange problem, almost as if when using IBL, Maxwell is using the FIRE engine automatically or something ...

[PA3K]

Hi ki_cz,

As I wrote before, but here is the simple explanation.
I can only attach pictures from maxwell render V3, but I am almost sure, that nothing will be different in V5.1.
As I wrote before, this is a "classic" - caustic through dielectrics problem.
You can see in my simple pictures how it works. Every "shadow" of transparent material is considered as caustic. Caustic can be refracted or reflected. Caustic can be seen directly or indirectly (refracted through another dielectric material or reflected through a mirror). The most difficult caustic is indirectly seen caustic of almost unidirectional beam light sources as is Sun (0,5°) or any small light sources shining from a big distance. There is nothing wrong with Your HDR images as You can see, there is good enough contrast between the shadow of a lit object and another unshadowed part. If there will be a problem with HDR images (dynamic range), there will be much less contrast or the edge of the shadow will be much softer. There is also no problem with the resolution of HDR as You can see.
You can see in the attached pictures, that the caustic is there, but in much higher sampling levels. Another point is that in the physical sun and sky scenario, there is a problem just with indirect seen refracted caustic. In the pictures with HDR, there is the same bad-looking directly (my surprise too!) and indirectly seen refracted caustic. And You can predict from small preview images, that the caustic will be there, but clean in SL 40 or even above. In the Fire engine, there is all caustic disabled (correct me please if I am wrong).

Patrik

[ki_cz]

Like, I suppose that I understand that the caustics are more complicated with IBL, but for a render engine to be unable to calculate a transparent single pane of glass in an IBL environment without reaching "SL 40" is a bit absurd to me.

I also find it weird that when using an MXI or HDR image on a spotlight or an emitter, then the caustics/transparency seems to work as it should. The problem seems to only exist when using an image for global illumination, which makes me think that there is something else going on behind the scenes other than it simply being too complicated of a calculation.

[ki_cz]

After a break I'm back to rendering and still wondering if anybody has any suggestions for how to deal with this issue? Simple single panes of glass taking until SL40 to be transparent with IBL seems kind of unacceptable to me, especially when it is so fast with emitters or physical sky ...

[Mark Bell]

Maybe start a fresh file and create some new AGS glass and go from there. I've had instances in the past where I've accidentally turned on/off a setting or adjusted a figure without realising it which has had a big impact on the render, but wasn't able to find what setting had changed for a long time.

[Artistus]

OMG! Now I understand the problem of darkening of the contents inside the glass bottle, like the cream inside the glass or huge amount of transparent grainy materials like salt, sugar etc. Because of this BUG (or maybe unsolvable issue), which takes A LOT more time to render caustics back through the dielectric it is impossible to use Maxwell for such projects. The render can take easily a week for 1 image for such a project. That's not serious at all.

There is single sheet of low grade glass 15*25 cm over the grey surface and 1*1 cm plane of light in attachment. If you try to render a camera lens this way, you'll get a lot of noise inside lens system. What I can see in the image below - it's ridiculous!

Maxwell team, please, check if this can be solved.

[ki_cz]

Thank you very much for adding some input Let's hope we hear something from NL soon, I agree that this makes a lot of projects impossible to complete.

[seghier]

Check this

https://blog.maxwellrender.com/tips/fast-and-frosted-2/

[Artistus]

Check this

https://blog.maxwellrender.com/tips/fast-and-frosted-2/

I'm afraid that this is not the solution for this kind of projects here.

I have tested the same scenario in Blender + LuxCore render.
1. With the default Path engine in LuxCore the situation is even worse, getting the result like fully dark shadow behind the glass sheet (HDRI map illumination in Maxwell).
2. When I change engine in LuxCore to Bidirectional and leave default Sobol sampler the issue looks the same as in Maxwell with physical sun or light plane for illumination.
3. But when I switch to Metropolis Light Transport sampler (which is available only in Bidir engine on CPU, the issue looks like to have optimization or some sort of workaround, so the caustics is getting back through the glass with a good grain of noise, which hopefully reduces MUCH MUCH faster than in any of previous options.

I'd like to have similar optimization available in Maxwell (like on 3rd example from LuxCore).

[ki_cz]

Thanks for the continued investigation, I will run a few more different tests once I have a chance as well, hopefully to help narrow down the issue for the NL team.

[Artistus]

Well, I have an update for this investigation. It seems like the issue is in the light size compared to the object size. When the light is relatively small (radial size from object's perspective), then it's very time consuming to get caustics from glass (it's shadow) back through that glass object. If the light source is relatively big, the issue magically disappears, but on the other hand caustics in this scenario will be less crisp.

You can compare the results in attachment. Expected result when light plane size is 10 cm, while sheet size is 5 cm. And odd result when light plane is 1 cm in size.

[ki_cz]

Um, that's an unbelievable find, I wonder if anybody from NL will chime in. This obviously seems like a bug at this point.

[ki_cz]

Hi all,

after some time I've returned to this and am still having the same issue. HDR Image Based Environment lights seem to just not work properly with glass. I'm attaching some additional tests.

Material is straight from Maxwell wizard, high-grade glass, no alterations, 1m x 1m x 1cm box, in a new 3ds max file, with no alterations to Maxwell settings except for dispersion and multilight enabled.

HDR Image Based Environment, dispersion on, SL13 (notice that not only does the light not pass through the glass, but you also don't see the reflected light):


Sun/sky based environment, dispersion on, SL13 (typical 'black' dielectrics in crossover sections), reflection of light works:


Normal Maxwell emitter (1cm x 1cm), dispersion on, SL13 (typical 'black' dielectrics in crossover sections), reflection of light works:


Maxwell emitter with HDR image (1cm x 1cm), dispersion on, SL10 (typical 'black' dielectrics in crossover sections), reflection of light works:


The fact that it works with an HDR scene emitter, but not the Image Based Environment, makes me think this is a problem within that section of the render engine? No other settings were changed.

Anybody have any idea how to avoid this? I prefer in most cases to use IBE for my work, but if it isn't compatible with glass, that's quite an issue.

[ki_cz]

One more test to add. This time with a created HDR in the Image Based Environment slot (a simple white light on black background), rendered to SL17. Still not even close to what I would expect from a scene with simple panes of glass, especially considering that it works far better in all other instances aside from Image Based Environment lighting. What I also find interesting is that in the preview image, the render looks as it should.



Hopefully somebody has a solution or ideas for why this simply doesn't work. Unfortunately I'm not really sure that while I love Maxwell, if I can't use HDR environment lighting for scenes with glass, if it's really going to be a consistent part of my workflow anymore.

[Artistus]

One more test to add...

Hi! It seems like it requires A LOT of rendering time when lighting the scene with HDRI which has small angular size of light in it. The same thing is more or less observed with regular Maxwell lights and sun/sky environment. I think when we look at the preview, it looks more like as it should, because it has a lot more samples per pixel for such a small image. On full size image we have separate dots of light in the shadows, while on the tiny preview those dots form solid area just because it's a lot smaller area there. I doubt that this "issue" is a really much of a fixable issue rather than annoying particularity of the rendering algorithm. Just because it has not only to emit light from the source, but check how much of it is emitted considering the hdr/opacity map applied. But maybe there some optimizations could be implemented that may slow down rendering speed in all other scenarios. I think (but not sure), it's not that easy, so that's why Maxwell team isn't participating in this thread.