Interior Lighting, Noise and DiffRadius Researching

[Yaroslav_L]

Hi everybody!
I am using Blender + YR / YaR for interior/exterior designing purposes many years.
And all this years I'm fighting with dirt, noise and render time.
In this post I want to share my Red Kitchen scene for researching.
I believe, together we can get nice, clear picture and correct lighting and rendering schema!
There are some important things in this experiment:
1. Scene scaling is 1 BU=1 meter! It's very important for DiffRadius.
2. The walls = 0.4 m! Not 1 cm, not 2 mm...
3. There are ALL walls! There wasn't front walls in some "interior" scenes, so that experiments are not true.
4. For real experiment we need real interior scene. Difficult to render engine scene. Much closet scene.
With different geometry and materials. The real, complex and detailed scene!
In this pictures I used my "Best" preset and only DiffRadius will changed.
Look at the tiles gloss reflection and noise under the table...

1. DiffRad=0.1


2. DiffRad=0.5


3. DiffRad=1.0


P.S. Unpack textures! And sorry for my EN.
http://rapidshare.com/files/296782641/K ... al-YaR.zip
MD5: 632CBB23BF992C1483714A6CFE5173F9

[victorphellipe]

The last image seems to be more realistic, but produce a white noise under the Window, near from desk. More AA settings will make it better. What settings do you using?
--
Nice render and nice topic!

[Arkinauta]

Many thanks for sharing your scene!

I´m learning PHOTON MAPPING now, so your scene just comes in a perfect time
I also didn´t know the importance of Diff Radius; I will follow your advices and tips, I found them quite handy. I also like the clean geometry of this interior.
And of course, the images are great

Thanks another time for sharing. Greetings!

[victorphellipe]

Arkinauta,
See this topic:
http://www.yafaray.org/community/forum/ ... =13&t=2690
It's a problem with diff radius, I think you'll understand better.

[xelptic]

I think i should hook up with this topic.. I'm also doing some experiments with PM, Diff radius etc.
So - time to download scene

ps. Yaroslav - could You upload scene on Mediafire or any other place instead of Rapidhshare? :/

[Yaroslav_L]

So share Your knowledge...

[xelptic]

As for now, i'm "rebuilding" some things in your scene:) For example materials: some of them uses fullbright colors (like red = 255, 0, 0) or adjusting lights power to don't have "overexposure" effect.
And modeling! I use somehow different approach to modeling walls and ceiling and floor. I usualy spend more time on modeling walls but i want them to be solid, closed, topologicaly correct models. In Your scene Yaroslav Ceiling connect with Walls in the way which is IMO incorrect. I mean they share only one edge instead of having "volume". Andsome faces of different models are overlaping.. I usualy correct this by let faces intersect or move them aside. I think correct modeling is one of the most important things when using PhotonMapping.

I'm sick now, but i'll try to do some tests to show you.

Best regards.
Xelptic

[Yaroslav_L]

Hi xelptic. Thanks for Your comments.
But I'm not fully agree with You.
For example colours: I need red, so because I use it, I don't need to think about details. As designer, I just choose the colour...
And about the way my walls and celling connected: Walls has correct dimensions and thickness and celling too. Normals has correct direction. I just put celling cube on the walls.
I left opened walls in another "rooms" for imitation some light coming from. I think for this scene (only kitchen modelling) there is no necessary to model windows in that another rooms. But when I designing all flat or house, of course I model all rooms (walls).
Everything are correct! And its work in another renders like old YafRay, VRay, kerkythea, modo's render...
But, it would be interesting to learn your modelling technique. Can You share some of your interior scene?
Thank You.
Sorry for my EN.

[bupla]

Correct me if I'm wrong Xelptic, but what he meant, is that there aren't pure colors in reality.
For example, if you need a white material in your scene (in a photoreal project), its color won't be 255, 255, 255, it should be a little less. It's the same for the saturation of the colors, like this red.

[Samo]

I believe is better if we don't consider Diff Radius value alone, but a as a combination of both Diff. Radius and Search. I mean, it is the number of photons found (Search) within a radius (Diff Radius).

If you decrease Search, there are less photons to mix therefore the likeliness to average a different result than the neighbour patch increases, hence the more noise.

Search 20
Diff Radius 1.0
Photon Count 200.000

20 search.png (131.99 KiB) Viewed 3918 times

If you increase Search, there are more photons in the mix, therefore the likeliness to average a different result than the neighbour patch decreases, hence the less noise.

Search 150
Diff Radius 1.0
Photon Count 200.000

150 search.png (104.83 KiB) Viewed 3918 times

Diff. Radius is the size of the photon map patch. If we decrease the size, we are increasing the signal frequency, there are more patches therefore is a more precise photonmap, but at the same time the noise produced by Search is increasing its frequency.

Search 20
Diff Radius 0.1
Photon Count 200.000

20 search Diff Radius 0.1.png (217.49 KiB) Viewed 3918 times

So better we use a good Search value. However, as we decrease Diff Radius, there are some patches that aren't getting enough photons for Search to mix, hence that group of patches will have got less luminance. An insuficient photon count will introduce low frequency noise again, even on a high resolution photon map.

Search 150
Diff Radius 0.1
Photon Count 200.000

150 search Diff Radius 0.1.png (150.67 KiB) Viewed 3918 times

Photon count should be increased as Diff. Radius decreases:

Search 150
Diff Radius 0.1
Photon Count 2.000.000

150 search Diff Radius 0.1 2Mphotons.png (170.62 KiB) Viewed 3918 times

Now that's a photon map ready for Final Gather to mix

[xelptic]

@Samo
That are important informations and test renders, thank you very much. It's quite similar to what I'm reading now in http://www.ypoart.com/tutorials/photon/index.php (Yves Poissant description how PM works and how to setup it in AnimationMaster - but almost everything can be transformed to Yafaray)

Also, I've found there some advices how to set DiffRadius and Search and Photon Count - in the mathematical way >> http://www.ypoart.com/tutorials/photon/sampling.php. First tests are quite satisfactory =]

Calculated DiffRadius test

Here is the scene: http://www.mediafire.com/download.php?lgomiyyhwn0

@Bupla - yes, I'm definitely sure we shouldn't use pure colors as diffuse color on materials Render above shows what's going on when we use Red=0.9,0.9,0.9 rgb >> unnatural color bleeding.

@Yaroslav - I'll post some more models soon

[Samo]

I see some estimation problems in that scene, particularly in some corners, can you post the blend file please

[xelptic]

It's a rough test - just put calculated radius into DiffRadius and render.

Scene posted. Have fun

[Samo]

Hi xelptic, this is the photon map of your previous scene:

That photon map has several issues which are likely to show up later with FG. For instance you should always watch for black spots, which means that the photon map isn't getting hit in that area, also with a relatively high photon count and a low diff radius the patches are still rather big and very incoherent in some areas.

If you dont' want smaller patches (faster) at least they should be rather homogeneous.

[Samo]

A better photon map, FG will be more comfortable with this:

Photon count: 5.000.000
Photon depth: 40
Diff Radius: 0.60
Search: 150

I also moved the arealight so its closer to the windows and all photons are accounted for, treat them well they are your little friends