Page 1 of 4

Interior GI test.

Unread postPosted: Thu Apr 04, 2013 1:48 pm
by Andz
I'm a part of a group o TD's and once in a while we have these small challenges where we have to test certain features on a specific scene on a myriad of rendering engines.

Right now, we are trying to calculate how much time we can gain on "preparing" a scene to render using brute force GI. So no ambient occlusion and direct light hack, no photons, etc. Ideally it should have a fast setup... usually given by brute force GI or pathtracers.

The first image which is cleaner I got from blender's Cycles, but it took 4 hours on GPU.
Image

This is the best I could achieve with Clarisse, I left it render over night.
Image

Can anyone give me advice on getting a cleaner render? Ideally in this challenge we were supposed to use only an dome/environment white light or a white emitting sphere, but in these two images I had to use area lights in each windows to reduce the noise. Issues with normals, faceting, etc. on the file are neglectable, we are only interested in the GI solution in this case

By the way, the cleanest/fastest image we have right now in the challenge came from Houdini. Around 24 minutes render time on my machine.

Thank you.
Andz

Re: Interior GI test.

Unread postPosted: Fri Apr 05, 2013 1:13 pm
by isoyann
Hi,

This kind of test is a very good one for any bruteforce GI raytracer... unfortunatly the PLE version have a bug in the secondary GI bounces (the problem start above 1 bounce.. the intensity drop way too quickly ), so you will not have optimal result with this build. I quickly rebuild a similar room to test this lighting setup, I get this result in 46 minutes on a dual Xeon E5645 @ 2.4Ghz...

Image
bedroom_clarisse_2778sec par isoyann, sur Flickr

I will test this night with a higher quality, to see if I can get a noise free result in 4H .
I used theses seetings : 1280x720
Raytracer AA samples : 8
Shading oversampling : 100% (this is not the recommanded value for general use... )
Two Area lights just behind each windows, quality : 3
1 GI light, quality :4 | Bounces : 3


How did you setup your render in clarisse? I'm also curious about the 20 minutes render in houdini... did it use brute force or any kind of interpolation (IR cache, point clouds... ) ? Is portal lights was used?

Thanks,

Yann Couderc
Technical Director
Isotropix

Re: Interior GI test.

Unread postPosted: Fri Apr 05, 2013 1:52 pm
by Andz
Hi Yann, thank you for the help.

Please disregard the 24 minutes Houdini rendering. I just checked the file and it is using photons for the indirect illumination and environment lights with sky portals. That is not a fair comparison. I'll try to look for the original file and I think that the pure PBR render was around 4 hours as well.

In case you would like to try with the scene with the original OBJ, the link is:
https://dl.dropbox.com/u/10937849/Bedroom.obj

I have reduced the object scale to 1% inside Clarisse.

My raytrace parameters are almost like yours:
AA: 8
Shading oversampling:100%
Area lights quality : 8
1 GI light: quality 4, bounces 3

Re: Interior GI test.

Unread postPosted: Fri Apr 05, 2013 5:00 pm
by isoyann
Thanks for the scene... I did a quick test with the same settings, here is the result :
Image
bedroom_clarissePLE_3133sec par isoyann, sur Flickr

Render time 3133 sec (52 minutes)
AA: 8
Shading oversampling:100%
Area lights quality : 8
1 GI light: quality 4, bounces 3
dual Xeon E5645 @ 2.4Ghz

Are you sure the rendering took one night? Clarisse don't use a progressive renderer, maybe the render was finished after one or more hours (depending on your PC), but not the full night? You can see the exact rendertime in the log (window/log).

Re: Interior GI test.

Unread postPosted: Fri Apr 05, 2013 5:18 pm
by Andz
Hi Yann, I meant that I left it render over night, not that it took the whole night.

Thank you for the test, I'll post your results for the rest of the team.

So, just to make things clear, your render was done with a newer (unreleased) build that fixes that secondary bounces issue, right?

Re: Interior GI test.

Unread postPosted: Fri Apr 05, 2013 6:39 pm
by isoyann
So, just to make things clear, your render was done with a newer (unreleased) build that fixes that secondary bounces issue, right?


This test was done with the regular PLE (build 5700) ... but the current release (available for all our clients under maintenance) have features and bugfixes not available in the current PLE. One of theses feature is the ability to texture the ray color exit of the GI.
Combined with distance clipping, you can get almost the same result in a fraction of the full GI rendertime, with less noise : here is an example render in 918sec (15 minutes) using this technique (the exit color is textured with a live 360 render of the room). The technique works perfectly in animation (no flickering), even with high contrast objects / lighting...

Image
bedroom_clarisse_918sec par isoyann, sur Flickr

Re: Interior GI test.

Unread postPosted: Fri Apr 05, 2013 7:19 pm
by Andz
"One of theses feature is the ability to texture the ray color exit of the GI."

Is that some type of light map? Final gather map?

Re: Interior GI test.

Unread postPosted: Fri Apr 05, 2013 11:05 pm
by isoyann
"One of theses feature is the ability to texture the ray color exit of the GI."

Is that some type of light map? Final gather map?


Not really.. using distance clipping, the GI light will only shoots rays in a max range of 1 meter for example (if you set the clipping to 1m). Above this distance, the rays hit nothing and return black. The ray color exit allows you to replace the black with any color, or texture it. Using a panoramic 360 small render (32x32 is enough) of your scene to texture it (in clarisse you can use any interactive render as a map) give you a very good approximation of the full GI render.
If you change the scene (move objects, add or change lights..) , the panoramic render will be re-rendered and used on the fly , here is an example which show this with a very saturated colored lighting, computed in 20 minutes (there is still noise in the 1 meter GI range because the sampling is too low, but the walls, illuminated mainly by the ray exit color, is clean.. ).

Image
bedroom_clarisse1200sec par isoyann, sur Flickr

Re: Interior GI test.

Unread postPosted: Wed Apr 10, 2013 10:28 am
by Makan
:) This Looks Awesome Keep It Up!

Re: Interior GI test.

Unread postPosted: Tue Jul 30, 2013 10:05 pm
by rawalanche
Hi... is the bug with secondary GI bounces still present in PLE release? I downloaded one just today, and folder still says build 5700b...

It has been a long while since this thread was created, it would be a real shame and not very good advertising if a bug that causes secondary GI bouncing problem was not fixed for such a long time. :(