PhotonGI cache

Discussion related to the Engine functionality, implementations and API.
User avatar
lacilaci
Donor
Donor
Posts: 1116
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Wed Mar 13, 2019 6:35 am

Ok, so switching to power(latest daily) works with many lightsources just fine, also with meshlights!
Not only that, but even simple enviro lighting gave me cleaner cache with power than with logpower.
I think if this problem cannot be fixed with logpower then power light strategy should be set as default, you rarely need only one lightsource in a scene :!:

User avatar
lacilaci
Donor
Donor
Posts: 1116
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Wed Mar 13, 2019 11:46 am

Another thing I noticed in few cases now, when rendering debug mode first pass is often mostly clean and right after that some weird noisy entries appear. What is that?

Screenshot from 2019-03-13 12-43-42.png

User avatar
Dade
Developer
Developer
Posts: 2531
Joined: Mon Dec 04, 2017 8:36 pm

Re: PhotonGI cache

Post by Dade » Wed Mar 13, 2019 11:57 am

lacilaci wrote:
Wed Mar 13, 2019 11:46 am
Another thing I noticed in few cases now, when rendering debug mode first pass is often mostly clean and right after that some weird noisy entries appear. What is that?
That is very strange, it is usually the result of hybrid rendering with CPU computing different results from GPU (so you have a blend of 2). Can you post an test scene ?
Support LuxCoreRender project with salts and bounties

User avatar
lacilaci
Donor
Donor
Posts: 1116
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Wed Mar 13, 2019 12:06 pm

Dade wrote:
Wed Mar 13, 2019 11:57 am
lacilaci wrote:
Wed Mar 13, 2019 11:46 am
Another thing I noticed in few cases now, when rendering debug mode first pass is often mostly clean and right after that some weird noisy entries appear. What is that?
That is very strange, it is usually the result of hybrid rendering with CPU computing different results from GPU (so you have a blend of 2). Can you post an test scene ?
I've sent you that scene in PM.
This scene also shows worse performance with gi caching than without (might be related)

At first I thought that it is happening cause the floor boards are very thin objects and maybe caches overlap/clip through, however it's on some walls too and I've seen this happening on other objects too. Everytime this happens performance also goes down to a point where it is worse than without gi cache!

EDIT: you're right, this does happen only in hybrid rendering - gpu only too.. cpu doesn't show that problem on gi cache debug!

User avatar
epilectrolytics
Donor
Donor
Posts: 402
Joined: Thu Oct 04, 2018 6:06 am

Re: PhotonGI cache

Post by epilectrolytics » Wed Mar 13, 2019 2:25 pm

Indirect cache is really great for difficult indirect lighting because the photons always find a way.
I made a test scene with a nearly closed door but there is a light leak issue.
Here is a reference (noisy) with BiDir and then the Path Sobol PGI version (same tone mapping value) where it looks like the bottom part of the door being transparent letting way too much light into the scene.
Did I do something wrong modeling the door or is it a PGI issue?
Rendered with latest builds as of march 13.
untitled1.jpg
untitled2.jpg
room3.blend.zip
(407.45 KiB) Downloaded 19 times
MBPro 15" 16GB i7-4850HQ GT750M, MacOS 10.13.6 & Win10Pro PC 16GB Ryzen 2700X, RTX 2070, GTX 1060

User avatar
lacilaci
Donor
Donor
Posts: 1116
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Wed Mar 13, 2019 3:43 pm

epilectrolytics wrote:
Wed Mar 13, 2019 2:25 pm
Indirect cache is really great for difficult indirect lighting because the photons always find a way.
I made a test scene with a nearly closed door but there is a light leak issue.
Here is a reference (noisy) with BiDir and then the Path Sobol PGI version (same tone mapping value) where it looks like the bottom part of the door being transparent letting way too much light into the scene.
Did I do something wrong modeling the door or is it a PGI issue?
Rendered with latest builds as of march 13.

untitled1.jpg

untitled2.jpg

room3.blend.zip
The biggest issue is that those few leaked samples are visible in reflection, you can raise glossiness threshold to get rid of that problem...
Screenshot from 2019-03-13 16-41-28.png
it won't be perfect, and I would definitely appreciate to have some ultimate solution for leaks, but it is what it is...

User avatar
epilectrolytics
Donor
Donor
Posts: 402
Joined: Thu Oct 04, 2018 6:06 am

Re: PhotonGI cache

Post by epilectrolytics » Wed Mar 13, 2019 3:49 pm

lacilaci wrote:
Wed Mar 13, 2019 3:43 pm
you can raise glossiness threshold to get rid of that problem...
Thanks!
That looks much better already though still different than the Bidir reference...
MBPro 15" 16GB i7-4850HQ GT750M, MacOS 10.13.6 & Win10Pro PC 16GB Ryzen 2700X, RTX 2070, GTX 1060

User avatar
lacilaci
Donor
Donor
Posts: 1116
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Wed Mar 13, 2019 3:55 pm

epilectrolytics wrote:
Wed Mar 13, 2019 3:49 pm
lacilaci wrote:
Wed Mar 13, 2019 3:43 pm
you can raise glossiness threshold to get rid of that problem...
Thanks!
That looks much better already though still different than the Bidir reference...
Cause the leaks are still there, and they carry a lot of power to illuminate interior. Look at left side of the door / hinge side where wall is, it is too bright cause a lot of cache entries are bouncing a lot of light on that wall, it's just that you don't see them in reflection, but they're still there...

User avatar
Dade
Developer
Developer
Posts: 2531
Joined: Mon Dec 04, 2017 8:36 pm

Re: PhotonGI cache

Post by Dade » Wed Mar 13, 2019 4:24 pm

epilectrolytics wrote:
Wed Mar 13, 2019 2:25 pm
Did I do something wrong modeling the door or is it a PGI issue?
The look up radius must be smaller of half of the door depth. So specify the radius by hand to a value like 1cm. To balance the very small radius you have also to scale back the brute force parameter (so from 6 to something like 6 * 15 = 90). And you get the correct result:

RGB_IMAGEPIPELINE_0.jpg
Support LuxCoreRender project with salts and bounties

User avatar
Dade
Developer
Developer
Posts: 2531
Joined: Mon Dec 04, 2017 8:36 pm

Re: PhotonGI cache

Post by Dade » Wed Mar 13, 2019 4:37 pm

BTW, in case it wasn't clear, more you reduce the radius, less bias you have and more it will look like BiDir rendering. This is a rendering with 25mm radius and 300 brute force scale:

RGB_IMAGEPIPELINE_0.jpg
Support LuxCoreRender project with salts and bounties

Post Reply