Test scences GPU Path vs CPU BiDir

Post your tests, experiments and unfinished renderings here.
Post Reply
User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Bubble Bubble

Post by FarbigeWelt » Sat Jul 21, 2018 11:27 am

There is stil a good distance to go until the visualization fits my imagination. But I made already a good part of the way. Here are some impressions of this hiking tour throu digital valleys and over binary hills.
Bubbles Glass Air Water CPU BiDir Metropolis.png
Bubbles Glass Air Water CPU BiDir Metropolis
Bubbles Glass Arc. Air Water CPU BiDir Metropolis.png
Bubbles Glass Arc. Air Water CPU BiDir Metropolis
Bubbles Glass Op. 01 Water CPU BiDir Metropolis.png
Bubbles Glass Opacity 0.1 Water CPU BiDir Metropolis.png
Bubbles Glass Water CPU BiDir Metropolis Rej. 2.png
Bubbles Glass Water CPU BiDir Metropolis Rej. 2
Bubbles Glass Water CPU BiDir Metropolis Rej. 1000 35 S 15 min.png
Bubbles Glass Water CPU BiDir Metropolis Rej. 1000 35 S 15 min.
Does anybody guess, what my imagination might be?

Rej. = Max. Consecutive Rejects
Arc. = Architecture
Op. = Opacity
Eye and Light Depth = 1000
Lamps = 2 ponints
Scene 79 MTriangles

EDIT:
How the bubbles got into their place.
https://youtu.be/y18u9_LfFBw
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Foam simulation is complex

Post by FarbigeWelt » Sun Jul 22, 2018 10:26 am

Sometimes my curiosity is much more intensive than any reasonable estimation. Do you ever had a closer look at foam? It seems to be white. But it is build from many colorless bubbles. The bubble is made from a thin liquid of water enclosing air. Parts are clear but foam looks white. Some people would call it a dispersive effect. But what is responsible for the dispersion? A single bubble is visible, it refelects and absorbs some light. Absorbance equals water and water is quite clear. If you stack bubbles and ignore reflections they should rahter be invisible. Even f you assume 1% energy lost of ray due to reflection you still need a lot of stacked bubbles to get full absorbance. Did you ever observed black foam in foam bath or on the surface of beer glass? I guess my set up is wrong because there is liquid in the space between aligned bubbles.
Beer Glass Water CPU BiDir Metropolis Rej. 2048 5500 S  12 h HD.png
Beer glass with something like foam
EDIT: Thanks to notes I've tried the approach with liquid between the air bubbles with a promising first result. By the way it renders 6 to 8 times faster.
Foam Sphere 667 kS per s 250 S Metropolis 19.8 M Tris.png
Foam Sphere 667 kS per s 250 S Metropolis 19.8 M Tris
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

wasd
Posts: 110
Joined: Tue Apr 24, 2018 7:20 pm

Re: Foam simulation is complex

Post by wasd » Sun Jul 22, 2018 1:28 pm

Bubbles and foam are a complex matter. Bubble is made of water captured between soap molecules. Walls thicknes is 10-1000 nanometers.
10.12691.wjce-4-2-2_20180722055607.png
White color is (in my opinion) is result of fresnel reflections.
depositphotos_2377391-stock-photo-soap-foam-bubble.jpg
A 3-dimensional foam is not a bunch of spherical bubbles, but a sctructure of packed bubbles. So, you'd need some means to create foam structure and a layered material, probably. Also note, that beer bubbles filled with carbon dioxide instead of air. It is slightly more refractive than air.
Foam_-_big.jpg
Upd:
I've managed to find a buch of scientifical papers on foam geometry. Well, I'm not so much of scientist, so studying for few years should get me to understand what's all that papers are about. Luckily, there's wonderful Surface Evolver program, that can save geometry to collada file. And the John Sullivan's Voronoi Diagram program, that can produce vertex data for the Surface Evolver.
Скриншот 2018-07-22 21.25.30.png
Last edited by wasd on Sun Jul 22, 2018 2:29 pm, edited 1 time in total.
CPU Bidir + Metropolis | Log Power | Core i5-4570

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Foam simulation is complex

Post by FarbigeWelt » Sun Jul 22, 2018 2:24 pm

Thank you for your input, wasd.

You are right it is even more complex than my simplification with spherical bubbles. And there are many different kind of foams in nature as your examples show well. How ever beer foam seems to be close to a compact packing of spheres because the bubbles are pretty and tiny. Soap bubbles are larger, their shapes are more complex if large enough and attachd to each other. There is a specific topology math which can be used to calculate organic looking robust roofs. Soap bubbles are very thin made from tenside and water. Changing thickness leads to the nice rainbow colors. Fresnel or another model might be used to render this colors. Would be a nice sim. I think soap bubbles are currently not possible with Blender-LuxCoreRender. Oligomolecular oil films on water lead also to nice, changing colors. Beer is made from e.g. water, ethanol, proteins, salts, hydrocarbons. As you mentioned beer bubbles are made from carbon dioxide. Densities of air gasses are roughly thousand times smaller than density of water. I guess the error for the rendered image is too small to be noticed. There are bigger influences on the look of the picture. In my imagination the white color of beer foam is mainly based on reflections and light conduction.

There are still some adjustments to be done for the final render. The room is too cold. The beer looks a bit thin. There are bubbles missing in the beer.
The foam is much to chaotic. Beer foam is quite regularly. How ever the foam improved a lot. But render speed is down to 150 kS/s. I have to check why. Maybe with environmental lightning sample count is much higher, could explain 667 kS/s for the last render.
LuxCoreRender counts 80 M Tris for the scene. Fireflies are again a matter, but a good part of them can be easy removed with Photoshop filter remove scratches, size 1.
Beer 150 kS per s 100 S Metropolis 80 M Tris.jpg
Beer 150 kS per s 100 S Metropolis 80 M Tris
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

wasd
Posts: 110
Joined: Tue Apr 24, 2018 7:20 pm

Re: Foam simulation is complex

Post by wasd » Sun Jul 22, 2018 2:42 pm

FarbigeWelt wrote:
Sun Jul 22, 2018 2:24 pm
How ever beer foam seems to be close to a compact packing of spheres because the bubbles are pretty and tiny.
I think it's because some bubbles collapse, spilling liquid between cells, so some kind of beer foam can be done with smaller bubbles between large bubbles and beer filling the gaps.
I currently don't have any beer to check it thouroghly.
FarbigeWelt wrote:
Sun Jul 22, 2018 2:24 pm
There are still some adjustments to be done for the final render. The room is too cold. The beer looks a bit thin. There are bubbles missing in the beer.
The foam is much to chaotic.
Render vodka! It has no bubbles, it is clear and nice.
Now it's time to check my reference and make sure it won't evaporate while I'm observing.
CPU Bidir + Metropolis | Log Power | Core i5-4570

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Foam simulation is complex

Post by FarbigeWelt » Sun Jul 22, 2018 4:01 pm

First I prerendered a minute or two to get the clamping guess. Then I checkmarked clamp using the suggestion. And started the render with Sobol adaptive set to 0.95. This lead to 215 kS/s for the 121 M Tris scence without any fire flies. Hurrah! Denoising is off.
The bubbles' size is a bit reduced, they have all the same size now. But sim's collection detecion is not perfect and most bubbles overlapping. I could pack more bubbles there but I like the few imperfections in the foam due to missing bubbles.

wasd wrote:
Sun Jul 22, 2018 2:42 pm
I currently don't have any beer to check it thouroghly.
Cheers! 8-)
Beer 214 kS per s 180 S Sobol 121 M Tris.png
Beer 214 kS per s 180 S Sobol 121 M Tris
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
B.Y.O.B.
Developer
Posts: 1825
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Test scences GPU Path vs CPU BiDir

Post by B.Y.O.B. » Sun Jul 22, 2018 5:09 pm

Suggestion: add drops on the surface of the glass, and get it to look better than the beer I did a few years ago: http://www.luxrender.net/forum/viewtopi ... =10#p93987 (probably not that hard :D )
Support LuxCoreRender project with salts and bounties

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Foam simulation is complex

Post by FarbigeWelt » Sun Jul 22, 2018 8:36 pm

Well, that is a good idea, B.Y.O.B.
Nice scence and your beer looks close to be drinkable.

At the moment I'm far away from a good scence. Table and chairs are old models. Some of the my first models actually.
Again, I'm fighting with fireflies using Sobol. Roughness seems to be one reason for them. A lamp between two double glassed windows is obviously not the best idea too to get a crisp sharp looking image. Metropoplis renders an interesting vintage look I did not expect. The light is a low polygon sphere with 2900 K. The walls are a simple matte material. And the beers' foams are red. For the second beer I had to run the particle simulation again. Okay, I could have converted them to meshs and join the spheres but the resulting mesh consumes to much ressources. I wonder how 485 MTris can be handeled.
Room with Table and Lamp 106 kSps 510 S 1h30 Metropolis.png
Room with Table and Lamp 106 kSps 510 S 1h30 Metropolis
EDIT: additional picture
Room with Table and Lamp 120 kSps 120 S 20 min Sobol.jpg
Room with Table and Lamp 120 kSps 120 S 20 min Sobol
In this case the expression "snow" make more sense. At least the foam is not red and the vintage look is gone. Probably a photography would look bad too in these circumstances. Supposed clipping value did not help much.

The beers need two things now, condense water to look cooler and a renderable scene. :-)
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Reasons for Snow: Ev. Light or Sun

Post by FarbigeWelt » Sun Jul 22, 2018 10:26 pm

Gain does not have any influence on environmental light's brightness.
Env. light has the ability to pass throu material matte and throu every little split in a model adding snow or fireflies.

I tried to have scene with env. light shining throu the window but no matter which gain I set, lamps inside the room had always been darker. I replaced env. light with a sun light. This lead to the snow (Sobol) and strange vintage look (Metropolis) in the pictures in from the last post.

I forgot about the sun light and assumed different reasons for the snow. I've replaced lamp and materials to avoid snow. Until I found the reason in the sun light outside the room.
Room with Table and Lamp 133 kSps 158 S 24 min Sobol.jpg
Room with Table and Lamp 133 kSps 158 S 24 min Sobol
This picture is close to what I expected from the setup.
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Just two intermediate Beers

Post by FarbigeWelt » Wed Jul 25, 2018 5:26 am

There are still a few things open. Water looks rather like wax. The water simulation is time consuming beeing improved. The foam of the beer in the back is too bright. The black background leads to strange looking water drops. And I have no idea why the beer looks turbid.
Two Beers.jpg
Two intermediate beers
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

Post Reply