PDA

View Full Version : 2019 Trial - first indoor BF GI test



madno
02-15-2019, 08:33 AM
Hi,

made a first GI indoor test with LW 2019 trial (VIP with no textures yet).
Intention was not to get the lowest possible render time with cheats and tricks but to see what the BF GI can do.

Settings for 7m34sec and 13m51sec are in the screenshot.

Change from 13m51sec to the next two (23min26sec and 42min29sec) was GI BF samples set to 50 and then 100.

All renders used new denoise.
Light is just a HDR in Textured Environment. No LW light.

With Portal MIS set to 700 samples (Portal light on the right) the HDR Env creates a nicely resolved shadow on the left wall.

As far as I remeber I was not able to get similar images of this scene with LW 2018 at a resonable time (I think I even tried hours of rendering).

All done on many years old dual Xeon (32 threads).
Every surface is native LW BSDF - no nodes touched.

The translucency on the curtain looks a little strange to me.


144153

7:34
144154

13:51
144155

23:26
144156

42:29
144157

Some comparisons at 249% zoom

1
144158

2
144159

3
144160

4
144161

vipvip242
02-15-2019, 10:04 AM
very interesting, thanks

jwiede
02-15-2019, 05:08 PM
The translucency on the curtain looks a little strange to me.

Agreed, it doesn't seem to be properly "passing through" color/light variations w.r.t. the view through the window, and instead just treating the entire view as if a uniform color & light level.

Is there a portal on that window by any chance? If so, does translucency behave the same way without the portal present?

madno
02-16-2019, 04:36 AM
Yep, there was a portal light. Below render shows the result when it is off.

144162

jwiede
02-16-2019, 02:19 PM
Yep, there was a portal light. Below render shows the result when it is off.

144162

Ugh, "damned if you do, damned if you don't" situation.

How are you manifesting the environment outside the windows? Is it an environment light? A backdrop of some sort? Or...? If not using env light, perhaps switch background to env light, omit portal, and see if that gives curtains properly translucent result?

Even if env light works around it, still seems worth filing a bug on the surface translucency / portal interaction in any case, pointing out translucency isn't accurately "receiving" color/light information from the portal (or if translucency should be, then portal isn't properly providing it), and including that scene as content.

RebelHill
02-16-2019, 04:44 PM
Curtain looks broadly ok to me, and theres no misbehaviour with portals and translucency either, illumination and colour carry through just fine.

Ma3rk
02-16-2019, 04:58 PM
I'm not so sure on the curtain guys. Seems believable to me. You've two different colors of light hitting from two different angles (look at the sills) and that's pretty much how I'd expect to see the darkening from the windw ledge. No sure quite what you'd call a bug anyway. I'd 'speriment a bit more first.

In any case, it's all very encouraging. I'm starting to get antsy to get my new system next month & delve into some of this stuff w/ a lot more horsepower.

RebelHill
02-16-2019, 05:09 PM
Yeah... I mean the curtain is perhaps the least realistic looking part, but I suspect taht this is because its just single poly thick... so infinitely thin. Ive seen this in many renders from many engines and its due to the fact that in the real world fabric does have some thickness, so the translucency is really SSS, there's internal self shadowing and all that stuff going on, which a single poly translucency shade doest quite capture.

madno
02-17-2019, 06:16 AM
Some more.

Flat double sided geo with translucency. Seems to be ok.
HDR sun comes from the right side mostly.

144164


Thickened geometry with subsurface.

144165

jwiede
02-17-2019, 02:02 PM
I'm not so sure on the curtain guys. Seems believable to me. You've two different colors of light hitting from two different angles (look at the sills) and that's pretty much how I'd expect to see the darkening from the windw ledge. No sure quite what you'd call a bug anyway. I'd 'speriment a bit more first.

It's being viewed through a thin, sheer curtain, not a thick piece of wax or plastic. Looking through the window, the tree is darker, that little "slice" where the lake shows through is substantially brighter, and then the ground below that is a little darker than the lake but still significantly lighter than the tree -- so the view's darker up top, then lighter "below" the tree's foliage. However, the curtain's showing none of that, it's as bright at the top where the tree is behind it in the view, as it is at the middle and bottom where the lake and ground are in the view.

File it, ignore it, whatever.

madno
02-17-2019, 04:36 PM
SSS on thickended geometry again. This time with other setting.

144166

madno
02-19-2019, 03:17 AM
After more testing I understand what jwiede means.

The SSS in BSDF material seems to be limited in what the scattering does. One would expect that a very thin surface with SSS should start to let the background shine through. I tried with thickness of 1mm, 0.3mm, 0.1mm but never got this "shine through. As it seems, one has to add transparency to the mix (= render time increases).

62 min. SSS plus transparency
144182

RebelHill
02-19-2019, 05:07 AM
One would expect that a very thin surface with SSS should start to let the background shine through.

Nope, thats transparency. Take a sheet of (preferably white) toilet paper, hold it up in front of your window, and see how much of the outside is visible through it (if you hold it right in from of your eye you can just about make out some basic shapes). Curtains like sheer nets give transparency effects because they're full of holes (often tiny but still there and making up a good portion of the total surface area). Drapes, or anything else with a tight weave doesn't because the light gets scattered, blurring the "image" to oblivion.

madno
02-19-2019, 07:07 AM
Yes, yes. Being to late to edit my post. Did exactly that. Looked at some thin objects around me and they behave the same like SSS in LW. So, transparency plus SSS or tranlucency in case of flat geometry are the way to go (if there would just not be the render time hit).

RebelHill
02-19-2019, 07:31 AM
Yep, do also note in a case like this, the transparency is because of tiny holes, not an actual transparent material... thus, there should be no refraction (ior 1)... so providing there are no other refractive objects in the scene, you can take the refraction samples down to 1, as any higher would (I think) incur a needless rendertime hit.