PDA

View Full Version : The new Clamped Intensity Falloff light options are nice - but....



Snosrap
10-24-2013, 12:22 PM
..... the F9 render doesn't match VPR.

Areyos Alektor
10-27-2013, 12:08 PM
It cannot be due to the draft mode ? (not tested yet)

Snosrap
10-27-2013, 03:19 PM
It cannot be due to the draft mode ? (not tested yet) Nope - with and without draft mode turned.

jwiede
10-27-2013, 06:24 PM
You have both set to the same colorspace, correct? The F9 looks really "hot" at that spot in that render, do you know what the fp numeric values are for the centers of those spots?

spherical
10-27-2013, 09:06 PM
I'm not sure how to set VPR to a different CS. However, with everything I can find set the same, I get this:

117870

All three have different color temperatures (more subtle in the screen grab than when looking at the real display).
The Render Status is cooler than the Image Viewer which is cooler than the VPR.
The Render Status is hotter than the Image Viewer which is hotter than the VPR.

Running a Levels on the screen grab on shadow, mid tones and highlight separately uncovers quite a bit, but this is on the screen grab. Being able to do the same on the actual display would show more of a difference that the eye can discern.

What I want to know is, what are the effective parameters of an Inverse Squared Clamped light? Benefits, as LW implements them. Searched the docs and web and came up empty. Pretty sure I get the concept from knowing physics but, as we all know, LW lights don't follow real-world. Seems like it's an attempt to do that, in some odd way, as opposed to making them really follow real-world physics and have relevant intensities that can be predicted; instead of 1,000% IES lights--which makes no sense.

I mean, what is the difference between a Range/Nominal Distance of 3 meters at 100% Intensity and a Range/Nominal Distance of 6 meters at 50%? Is there one? Before you pick at the math involved, I now all that, and I know that it also depends upon the falloff type chosen. This is meant to be a simple example. I'm sure that the concept is clear.

jwiede
10-28-2013, 03:25 AM
What I want to know is, what are the effective parameters of an Inverse Squared Clamped light? Benefits, as LW implements them. Searched the docs and web and came up empty.

I suspect this is the central question. Someone from LW3DG needs to explain what they _intended_ from these falloff settings? Without a quantitative explanation of how the clamping is supposed to work, there's no way to judge functionality.

They certainly do not seem to work in the way a user would rationally expect them to work.

Lightwolf
10-28-2013, 05:25 AM
I suspect this is the central question. Someone from LW3DG needs to explain what they _intended_ from these falloff settings? Without a quantitative explanation of how the clamping is supposed to work, there's no way to judge functionality.

It's to prevent overly hot areas close to the light. Without the option, the light intensity can ramp up tremendously if the distance to the light is less than the nominal range.

Cheers,
Mike

Snosrap
10-28-2013, 08:26 AM
It's to prevent overly hot areas close to the light. Without the option, the light intensity can ramp up tremendously if the distance to the light is less than the nominal range.

Cheers,
Mike

Yes - exactly. And it works and looks great in the VPR but renders hot with F9.

Lewis
10-28-2013, 08:53 AM
Yes - exactly. And it works and looks great in the VPR but renders hot with F9.

I've reported difference/not working in VPR and Fogbugz respond said it'll be added to VPR features in next build(s). So it seems they were already aware of that :).

allabulle
10-28-2013, 11:20 AM
Good to know, thanks Lewis.

jwiede
10-28-2013, 12:47 PM
It's to prevent overly hot areas close to the light. Without the option, the light intensity can ramp up tremendously if the distance to the light is less than the nominal range.

Right, that's the obvious qualitative explanation, but what is the expected clamping threshold? Is it just whatever the light puts down at nominal distance, or is it more elaborate than that?

Snosrap
10-28-2013, 04:10 PM
I've reported difference/not working in VPR and Fogbugz respond said it'll be added to VPR features in next build(s). So it seems they were already aware of that :).

Well for me it seems to work with VPR. It's an F9 that is not being clamped. :)

spherical
10-28-2013, 08:19 PM
It's to prevent overly hot areas close to the light. Without the option, the light intensity can ramp up tremendously if the distance to the light is less than the nominal range.

Cheers,
Mike

Ok, but I thought that the 11.x series already handled this better. That was one of the big selling points, IIRC.

Still need to grok the Nominal Distance/Intensity thing. Anyone?

pinkmouse
10-29-2013, 02:18 AM
I wonder if Ben's written that part of the new manual yet? :)

probiner
10-29-2013, 04:57 AM
Good feature to avoid white artifacts usually like a cross, from using Inverse Squared lights.

BeeVee
10-29-2013, 07:04 AM
I have and it's got pictures too! :D

B

Lewis
10-29-2013, 07:20 AM
I have and it's got pictures too! :D

B

As they say:
117887

lardbros
10-29-2013, 07:44 AM
Like the original poster says, it definitely looks like F9 isn't working properly, but the VPR is... nice feature by the way, this will really come in handy.

Does it help with overly bright reflection aliasing issues too? :D

Tobian
10-29-2013, 07:59 AM
The feature basically clamps the intensity at the range/nominal distance. So say you have an intensity of 5000 at a range of 10cm, then it will get no brighter than 5000 at a range of 10cm... For normal inverse squared falloff, it goes up in intensity by the same inverse squared intensity ramp, so in the above example the intensity at 1cm might be a few million % bright (without doing the maths), but over an incredibly small area. This will create fireflies, so the idea is to stop that from happening.

It will help in any situation where said lights intersect with geometry, be it illuminating diffuse geometry, specular hotspots, or reflections of such things.

BeeVee
10-29-2013, 08:27 AM
As they say:
117887

https://dl.dropboxusercontent.com/u/2807104/forum/clamp.png

B

spherical
10-29-2013, 04:03 PM
The feature basically clamps the intensity at the range/nominal distance. So say you have an intensity of 5000 at a range of 10cm, then it will get no brighter than 5000 at a range of 10cm... For normal inverse squared falloff, it goes up in intensity by the same inverse squared intensity ramp, so in the above example the intensity at 1cm might be a few million % bright (without doing the maths), but over an incredibly small area. This will create fireflies, so the idea is to stop that from happening.

It will help in any situation where said lights intersect with geometry, be it illuminating diffuse geometry, specular hotspots, or reflections of such things.

SOOOOoooo, is this closer to actual physics or farther away from it?

Lightwolf
10-29-2013, 04:13 PM
SOOOOoooo, is this closer to actual physics or farther away from it?
Neither, depending on how you use it. ;)

Cheers,
Mike

spherical
10-29-2013, 04:30 PM
Well, there is that...

Lightwolf
10-29-2013, 04:37 PM
Well, there is that...

:D

Well, on the one side, the idea of the light intensity ramping up to infinity, as it did before, is wrong. On the other hand, the nominal distance is kind of wrong either way.

What you could do with the current system is set that to the distance of the (virtual) light emitting surface (if you visualise the light as a "real" light) and then assume the intensity to be the intensity that is emitted. That would be fairly correct in physical terms (except for light with a different falloff characteristic).

But, except for the ramp to infinity, it's something you could have done earlier as well. I just never found it to be very practical when lighting.

If it looks right it's right.

Cheers,
Mike

spherical
10-29-2013, 06:45 PM
Well, on the one side, the idea of the light intensity ramping up to infinity, as it did before, is wrong. On the other hand, the nominal distance is kind of wrong either way.

Which is my take on it, yes.


What you could do with the current system is set that to the distance of the (virtual) light emitting surface (if you visualise the light as a "real" light) and then assume the intensity to be the intensity that is emitted. That would be fairly correct in physical terms (except for light with a different falloff characteristic).

So for a bulb style, you'd set the Nominal Distance to the radius of the glass envelope or the filament itself? I suppose that if the envelope is frosted, the glass radius would be the emitting surface. In the case of a halogen, the envelope is darn close to the filament and the envelope is normally clear.

From the manual:
"the Range/Nominal Distance value determines the point where the Light Intensity reaches the Light Intensity setting"
is a bit ambiguous. This does seem to say what you described. However, I thought I remembered reading something along the lines of:

"the Range/Nominal Distance is the distance at which the beam intensity drops to 50% of the Intensity Setting"
but cannot find that written in manuals back to v9 now.


But, except for the ramp to infinity, it's something you could have done earlier as well. I just never found it to be very practical when lighting.

If it looks right it's right.

I've been attempting to get a simulation of what a light fixture's beam throw would be and, with this method of complete fakery, it is a useless endeavor. Everything else is focused on being as real as is possible but the lights are just ignored on that front altogether. Doesn't make any sense. What Lewis has observed with IES lights, which I thought would be far better on the realism scale than standard LW lights, just boggles the mind.

There really needs to be a standard of some sort that one may base lighting against. I don't care if it's a function that is internal to LW and we never access it. Just make 100% on a 50W IES light emulate that light at its full voltage. The beam pattern is there. The intensities within the beam are there. All that's missing is a relative brightness based upon a known standard like luminous flux.

Lacking that, if there were a calibration tool where a modeler could dial in a chosen light in the scene against a measuring scale of some sort (similar to a basic visual gamma calibration) and then base all other light intensities in the scene against that one would work.

Snosrap
10-29-2013, 10:40 PM
In the new build released today (build 2721 release candidate 4) clamped lights don't work at all in either VPR or F9.

Lewis
10-30-2013, 01:20 AM
In the new build released today (build 2721 release candidate 4) clamped lights don't work at all in either VPR or F9.

Hmm, are you sure you are using it right i.e. understand what it does ? I've just tested 2721 and it works fine in both VPR and F9 here.

use Area light and move it close to the ground and see (check the hotspots/fireflyes on area close to ground):

spherical
10-30-2013, 02:03 AM
OK, so the preferred choice is clamped and non-clamped is retained presumably for backward compatibility. Haven't noticed a detectable render time difference between the two, but my test was in the sub-second range, so....

BeeVee
10-30-2013, 07:28 AM
Not necessarily just for backward compatibility. The look might be something you'd still want. Because of the way clamping works you might not want to use it since you can end up with a large area of monotone.

B

Snosrap
10-30-2013, 08:10 AM
Hmm, are you sure you are using it right i.e. understand what it does ?

use Area light and move it close to the ground and see (check the hotspots/fireflyes on area close to ground): Yes, I think. :) - there is no discernable clamping happening in the new build (2721) between unclamped and clamped lights. Here are two screen shots of VPR in build 2717 and build 2721, notice the difference with clamping with and without in build 2717 versus no difference in build 2721. I'm including the scene for your own tests. As in BeeVee's screeshot clamping should tone down the hotspot, your hotspot looks the same - the fireflys are most likely sampling issues or other. Anyway I think this needs further testing and user education as I may not know what the %^& I'm talking about. :)

Lewis
10-30-2013, 08:15 AM
your hotspot looks the same - the fireflys are most likely sampling issues or other.

Not at all, my sampling/AA is SAME in both renders and clamped one don't have firefly's 'coz it's clamped i.e. light don't go over 100% intensity (to infinite ;)) while in non clamped it's extremely bright in that close area and that's why i got firefly's. Move the Area light closer to ground in your scene and you'll see it working just fine. It doesn't need to tone down lighting to be "fine".

EDIT: Also don't expect Luminous geometry to be clamped, you are using luminous poly and not just light.

Lewis
10-30-2013, 08:25 AM
Here is your scene adjusted to see difference clearly.

do you see it now ? Ground is much better and reflection also 'coz there is no overblown pixels/no firefly's

probiner
10-30-2013, 08:36 AM
Just a trivia, in Arnold there is no nominal distance control control input and there's also no choice for the falloff type it's always inverse square distance.

Tobian
10-30-2013, 08:41 AM
NB: there was an initial bug with the appearance of the clamping in VPR vs F9 which has been fixed. The point is the clamping is inside the falloff radius, so if there is not much geometry inside this radius, the chances are you won't see much difference.

BeeVee
10-30-2013, 10:00 AM
Try moving the area light further away from the scene and increase the falloff so the objects are still inside it, then see the difference.

B

spherical
10-30-2013, 11:44 PM
Just a trivia, in Arnold there is no nominal distance control control input and there's also no choice for the falloff type it's always inverse square distance.

Exactly my point. Inverse square should BE inverse square. It should just follow real-world.


Try moving the area light further away from the scene and increase the falloff so the objects are still inside it, then see the difference.

B

Umm, okay, why do we have to go through these machinations?

BeeVee
10-31-2013, 04:49 AM
You don't have to, it was just to exaggerate the effect to make it more visible.

B

Lightwolf
10-31-2013, 08:37 AM
... and there's also no choice for the falloff type it's always inverse square distance.
I hope not always, as there's cases where it's not the right choice.