PDA

View Full Version : Help with Dielectric node



Matt Steele
02-21-2014, 06:06 PM
Hey guys,

Long time Lightwave user, (over 20 years) and I have become ridiculously frustrated with our glass surface we use at work. I was wondering if anyone might be able to help shed some light on what is causing the problem.

I'm attaching a surface preset scene we have made to help us with our surfaces. The issue I can't seem to resolve is how to get our dielectric node glass surface to render properly and appear transparent (not turn black) when laying on top of another surface.

120296

As you can see, the surface turns black when we want to be able to see through the surface to the ball underneath. I've seen many other scenes and surfaces render this type of image properly. What's strange is that the surrounding glass surface looks good, but the surface above the ball does not.

The scene is lit through an HDRI image and global illumination.

Any suggestions would be very much appreciated. Thanks!

Lightwolf
02-21-2014, 06:42 PM
How many recursions are you rendering with?

Cheers,
Mike

spherical
02-21-2014, 07:27 PM
Try getting a bit of air between the surfaces. Just make the polys non-coincident by the tiniest amount. Unless the surfaces that are in contact with each other are both transparent, like water in a glass where there would be no air between, many times you'll get this.

Matt Steele
02-21-2014, 08:59 PM
How many recursions are you rendering with?

Cheers,
Mike

Thanks for your reply Mike. I do not recall what the scene at work is at, but the one I just re-built here at home is set to 6. If I raise it any higher, I get really bright lines within the edge bevel of the glass which is really not good looking at all.


Try getting a bit of air between the surfaces. Just make the polys non-coincident by the tiniest amount. Unless the surfaces that are in contact with each other are both transparent, like water in a glass where there would be no air between, many times you'll get this.

Thanks for the reply spherical, there is a gap between the ball and the glass. We even tried eliminating the inside portion of the glass to see if that would make a positive difference, and it worsened the problem.

I've re-built a similar model here at home tonight and am getting better results with a completely different lighting set up. It's not ideal since I'd like to be using lighting similar to what we will be using within our projects, but the glass dieletric is improved. Although now the beveled edges are getting very dark. :(

120299

spherical
02-22-2014, 12:24 AM
I do not recall what the scene at work is at, but the one I just re-built here at home is set to 6. If I raise it any higher, I get really bright lines within the edge bevel of the glass which is really not good looking at all.

This doesn't make sense. The higher the RRL, the better it should look; up to a point of useless recursions that don't do anything but waste time. So, this says that perhaps the expectation of how it should look is not matching how it does look.


There is a gap between the ball and the glass. We even tried eliminating the inside portion of the glass to see if that would make a positive difference, and it worsened the problem.

Dielectric requires a closed volume, so creating a false open structure will definitely screw things up. If the volume isn't closed, Total and Partial Internal Reflections cannot work.


I've re-built a similar model here at home tonight and am getting better results with a completely different lighting set up. It's not ideal since I'd like to be using lighting similar to what we will be using within our projects, but the glass dieletric is improved.

Attaching scenes/models would help greatly in debugging this. In this way, we can see how things are being set and provide better solutions.


Although now the beveled edges are getting very dark.

This sounds like RRL again. I have had some models that required odd or even numbers of RRL in order to have the edge appear light or dark, as I expected them to be. IOW, if the bounces end up going away from the camera as the ray exits the glass volume, you get a dark edge. Raising or lowering the RRL value by 1 makes the edge bright. In the Real World, there is no ray limit.

EDIT: Just took a look at the image provided and the bright areas that you don't like are to be expected. They are not out of line, they just don't match that which you expect.

Matt Steele
02-22-2014, 11:18 AM
This doesn't make sense. The higher the RRL, the better it should look; up to a point of useless recursions that don't do anything but waste time. So, this says that perhaps the expectation of how it should look is not matching how it does look.



Dielectric requires a closed volume, so creating a false open structure will definitely screw things up. If the volume isn't closed, Total and Partial Internal Reflections cannot work.



Attaching scenes/models would help greatly in debugging this. In this way, we can see how things are being set and provide better solutions.



This sounds like RRL again. I have had some models that required odd or even numbers of RRL in order to have the edge appear light or dark, as I expected them to be. IOW, if the bounces end up going away from the camera as the ray exits the glass volume, you get a dark edge. Raising or lowering the RRL value by 1 makes the edge bright. In the Real World, there is no ray limit.

EDIT: Just took a look at the image provided and the bright areas that you don't like are to be expected. They are not out of line, they just don't match that which you expect.

Spherical, thanks for your response. I've rendered another rendering with the RRL at 64. To me, this is beginning to look quite unrealistic. What's your opinion?

120374

The glass object is a closed volume. I was just mentioning that we tried opening it up just for the sake of seeing what would happen in an attempt to correct our problem, and it did make the situation even worse. Never know till you try right? :)

I'm going to continue to work on this, but I do agree that the RRL may help resolve some of the issues. Thanks for your suggestions!

spherical
02-22-2014, 03:47 PM
Looks great. What do you think is unrealistic? There appears to be some possible noise in the internal reflections but, at this scale, I am unable to tell if they are actual recursions or just noise from the sampling levels chosen. Might turn on Limit Dynamic Range and see if they clear any.

What are your Shading Samples, Light Samples, GI settings and AA/AS settings?
What does the lighting HDR look like and is it blurred?

One way to tell if it is recursions is to construct a debug HDR having stripes of different colors or leave that alone and place different color objects around the glass object and see where the colors show up.

XswampyX
02-22-2014, 05:18 PM
You don't want to be running RRL @ 64 for this. It's just a glass sphere covering a ball!

Matt Steele
02-22-2014, 07:48 PM
Looks great. What do you think is unrealistic? There appears to be some possible noise in the internal reflections but, at this scale, I am unable to tell if they are actual recursions or just noise from the sampling levels chosen. Might turn on Limit Dynamic Range and see if they clear any.

What are your Shading Samples, Light Samples, GI settings and AA/AS settings?
What does the lighting HDR look like and is it blurred?

One way to tell if it is recursions is to construct a debug HDR having stripes of different colors or leave that alone and place different color objects around the glass object and see where the colors show up.

Glass does not look like that!The render is totally unrealistic.

Shading samples are 8 and light are 16. We have found these settings are more than acceptable for nice smooth renders with very low noise. 8X anti aliasing and the scene GI was just the Preset Central basic scene. One would think those settings should be perfect for what I'm attempting to do... Set up a scene to develop basic surfaces with.


You don't want to be running RRL @ 64 for this. It's just a glass sphere covering a ball!

Agreed, not to mention it makes the render appear horrible. I was willing to attempt anything to try to resolve this issue.

Matt Steele
02-22-2014, 09:03 PM
Here's another attempt with RRL set back to 6. I'm pretty happy with this...

120383

spherical
02-22-2014, 09:21 PM
Well, there ya go...

I don't see any call to get cranky with me. I'm trying to help you. You asked my opinion; I gave it. You never posted the files that I asked for, so I could see how you have things set. Did you try any of the suggestions I made? Do you have any absorption set? Once you came back with something, I was going to suggest cutting RRL by half in multiple steps to arrive at a workable and fast compromise.

Look, Dielectric is my thing. I make objects from hot glass that is heated to 2,280F. I know what glass is supposed to look like in all of its states. Using that experience, I helped develop Dielectric by filing many FogBugz reports, based upon many tests and comparisons. If you've found some problem that needs addressing because, "Glass doesn't look like that!", when I am comparing results to real-world crystal and flint glass examples sitting on my desk, I'd really like to know the specific details; and so would the developers. Subjective conclusions are just confusing.

Granted, Dielectric in LightWave still has a ways to go in some areas, like caustics, when compared to an unbiased renderer (outlined in another thread dedicated to Thea Render and their differences). Still, it's pretty darn good, and fast to boot. I've run it through the wringer and, when it comes to glass, I'm pretty darn picky.

spherical
02-22-2014, 09:25 PM
This stupid forum... The ol' 8:20 pm PST NewTek forum hang gets me again, this time with a double post.

Matt Steele
02-22-2014, 10:03 PM
Well, there ya go...

I don't see any call to get cranky with me. I'm trying to help you. You asked my opinion; I gave it. You never posted the files that I asked for, so I could see how you have things set. Did you try any of the suggestions I made? Do you have any absorption set? Once you came back with something, I was going to suggest cutting RRL by half in multiple steps to arrive at a workable and fast compromise.

Look, Dielectric is my thing. I make objects from hot glass that is heated to 2,280F. I know what glass is supposed to look like in all of its states. Using that experience, I helped develop Dielectric by filing many FogBugz reports, based upon many tests and comparisons. If you've found some problem that needs addressing because, "Glass doesn't look like that!", when I am comparing results to real-world crystal and flint glass examples sitting on my desk, I'd really like to know the specific details; and so would the developers. Subjective conclusions are just confusing.

Granted, Dielectric in LightWave still has a ways to go in some areas, like caustics, when compared to an unbiased renderer (outlined in another thread dedicated to Thea Render and their differences). Still, it's pretty darn good, and fast to boot. I've run it through the wringer and, when it comes to glass, I'm pretty darn picky.

spherical, sorry if you took my post as me being cranky with you. I appreciate the help. I was just reacting to what I thought obviously did not look good. But beauty is in the eye of the beholder so to each their own. You obviously know more about dielectric surfaces than I do. You probably have studied them more than I have and it's possible the harsh reflections/refractions I was disliking in the bevel areas of the shape might be realistic. They just did not seem so to me. Again, thanks for the suggestions. Much appreciated.

spherical
02-22-2014, 10:08 PM
Thank you for your explanation. Yes... beauty, or lack thereof, is in the beholder's eye. It is when we are confronted with a vision that we do not expect is when the challenge commences to learn which is actual.

Still would like to learn what all of your settings were.

Matt Steele
02-22-2014, 11:30 PM
Still would like to learn what all of your settings were.

I just pulled down the Preset Central scene file, changed a few rendering properties, light/shade samples/anti aliasing, and left everything else alone. Added in the hdri image for the background and reflections and everything else was tweaking the settings on the glass dielectric. That's it. I changed the RRL many times as I said but ended up with a setting of 6.

spherical
02-23-2014, 01:24 AM
OK, which PCCentral scene, there are 12 of them, set for different scales, and that isn't their standard test object.

Did you set Absorption? Using RRL of 64 with zero or near-zero Absorption would cause odd things after so many recursions. That is what I'm trying to learn. There are a lot of things that could affect a glass render.

Matt Steele
02-23-2014, 07:40 AM
OK, which PCCentral scene, there are 12 of them, set for different scales, and that isn't their standard test object.

Did you set Absorption? Using RRL of 64 with zero or near-zero Absorption would cause odd things after so many recursions. That is what I'm trying to learn. There are a lot of things that could affect a glass render.

The scene I did at work was completely custom. The one I re-created at home this weekend was from the preset central 10cm area lights scene. Yes, the model is custom to resemble the maxwell renderer surface tester. The glass surface started out as being completely white with 4.0 absorption. I ended up making it a grey with 1.0 absorption. When the glass surface is white, higher rrl settings produce undesirable effects (hot white lines within the glass reflections).

spherical
02-23-2014, 03:01 PM
Yep, that would be consistent. Without having your simball, and not knowing how it is constructed, I couldn't debug further so was left to logic and having been there. I'll work some today with the 10cm area. Thanks for the info.