GTR and animation fireflies

[^Lele^]

The relentless drive for excellence, right there.
Is it safe to say we have the only implementation of the full GTR domain on the market?

[Algehov]

Thats great to hear Vlado!

Lele, what you said sounds very interesting, but I struggle to understand how glass for example rarely goes above 15% in reflectance amount. When I look at a glass window at a very narrow angle I can see it reflect the environment at almost full "intensity". Or mirror like. Do you mean, if measured that would actually only be 15% reflected back to me? Cause it looks very close to 100%.
Unless you're talking about dead on 0 degree vieweing, but that should be handleded by the fresnel and not the reflection amount right?

Or more likely Im just not understanding you right. But I would love to hear more about that!

I've had a feeling that reflections in many renders are too weak compared to reality (expecially glass windows on exteriors). And had a theory on maybe it had to do that our diffuse or albedo values are normaly set to high compared to what a measured value would give us. Which in terms would result in us under exposuring the image to compensate for the brighter albedo. And then getting weaker reflections (atleast weaker reflections from lights).

But it sounds like most materials might have a darker reflection than you would expect, aswell as a darker albedo?
Sorry if im rambeling

Thanks in advance,
Simon

[^Lele^]

Lele, what you said sounds very interesting, but I struggle to understand how glass for example rarely goes above 15% in reflectance amount. When I look at a glass window at a very narrow angle I can see it reflect the environment at almost full "intensity". Or mirror like. Do you mean, if measured that would actually only be 15% reflected back to me? Cause it looks very close to 100%.

Yes, i meant the maximum reflectance.
The very different perception is chiefly down to the human visual system (that includes the brain) being non-linear, with a bias towards boosting the "darkness" we live in.
But there are a few simple experiments one can make to find out, and a few are even just thought experiments.

The simplest one i can think of off the top of my head is this: have you ever seen reflective caustics off the face of glass buildings, making pools of light in otherwise shader areas? It's a very common occurrence in cities like Vancouver, with their city centre being covered in high rise buildings of very modern make.
Even accounting for off-angles and other dimming effects, those light pools are superbly faint, when compared to the edge of the shadowed area, where it meets direct light on the concrete or tarmac pavement.
And if you turned to look at the sun (so with just the atmosphere between it and your eyes) you'd positively go blind.
Simpler still would be to use a set of polarised glasses to look at the brightest of window reflections you can, reflecting the sun directly: the reflection will be dimmed quite a lot, likely becoming bearable to watch.
Which isn't going to happen if you then stare back at the sun: polarisation or not, it's just too bright a source to be dimmed by a thin glass lens.
Now, i'm not saying there aren't 90%+ reflective materials in nature, mind you, it's just that us "normal" humans hardly ever get to experience those, with a run-of-the-mill bedroom mirror sitting comfortably at around 80 to 90% total reflectance, much like a carpaint clearcoat at extreme grazing angles.

Which takes us to the point i am trying to make (well, that i tried to make for the past nine years, starting with a tutorial on using the sun and sky, back in 2007).
PBR needs the combination of three factors to produce a correct rendering: shaders albedo, lights intensities, and exposure.
More often than not, the equation (the above would look like PBR= S * L * E) falls apart at the shader level, while lights (IES profiles and real units help with that) are right, and perhaps also exposure would be, IF the shaders were right too.
But if the shaders are too bright (in other words, their total albedo too high), the lights will not dim upon bouncing on them, generally brightening the results (and in cases leading to fireflies), which then the user compensates for by tweaking exposure, or bouncing back and forth between exposure and shaders, or soft and hard clamping the lighting equation or the image.

Soon enough we'll have the VRscans scenes out for everyone to try out, and doing control renders on those shaders will provide with exact, measured reflectance data.
Below a sphere rendering under 1.0f dome lighting and then with an HDRI of, respectively, white satin cloth and a blue carpaint (without clearcoat), rendered with no exposure control.

[Donfarese]

All this info is great to read, and what you guys are doing with vray is awesome, just keeps getting better and better. I've noticed this too lately, more fire fly like noise on objects with glossy GGX surfaces. I don't remember it happening before or just never noticed it, But i do a lot of tests so I'm surprised I missed it. I attached a very basic scene, I noticed very slightly like fireflies on the top of the pot and a couple on the sphere. This is very slightly, I had to turn up the AA and Shader Quality to get rid of them, but they still kinda remain. Seems to take long to clear, especially compared to blinn. Almost looks like dust particles. Take a look if this helps, I attached the scene, also try the same render with blinn and you'll see the difference.

[^Lele^]

All this info is great to read, and what you guys are doing with vray is awesome, just keeps getting better and better. I've noticed this too lately, more fire fly like noise on objects with glossy GGX surfaces. I don't remember it happening before or just never noticed it, But i do a lot of tests so I'm surprised I missed it. I attached a very basic scene, I noticed very slightly like fireflies on the top of the pot and a couple on the sphere. This is very slightly, I had to turn up the AA and Shader Quality to get rid of them, but they still kinda remain. Seems to take long to clear, especially compared to blinn. Almost looks like dust particles. Take a look if this helps, I attached the scene, also try the same render with blinn and you'll see the difference.

http://forums.chaosgroup.com/showthr...531#post690531

As for clearing them, they will, just let it sample (instead of stopping it forcefully at 6 subdivs.)

[Donfarese]

http://forums.chaosgroup.com/showthr...531#post690531

As for clearing them, they will, just let it sample (instead of stopping it forcefully at 6 subdivs.)

Will do bud thanks for all the info.

[vlado]

Just for comparison, here are the result between our original GGX sampler, and the new improved one in V-Ray 3.40, using the example from this thread:

Old GGX sampler:


New GGX sampler:


Best regards,
Vlado

[bardo]

Wow!
Nice improvement!!!

You rock guys!

[Donfarese]

Man that is an Improvement! awesome work Vlado.

[Algehov]

Holy shit Vlado. Thats some great results!

@Lele
Ye, having some nice vrayscan samples would be a great reference for this kind of stuff! Thanks for the reply. Interesting for sure

[^Lele^]

Yep, i got tasked with getting an albedo table off the scans library.
Weekend made!