PhotonGI cache

Discussion related to the Engine functionality, implementations and API.
User avatar
lacilaci
Donor
Donor
Posts: 1381
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Tue Feb 12, 2019 11:34 am

Also, trying to use photonGI in a really heavy scene (exterior + ton of high poly trees) crashes blender with error:
Screenshot from 2019-02-12 12-33-25.png
here's how the scene that crashes with photongi looks like (bidir+metro)
scene.jpg
hmm.. there's probably some problem with the scene itself, not luxcore related.
EDIT: it was some issue with the scene, I've reimported trees in another copy of that scene and caching works just fine!
Last edited by lacilaci on Tue Feb 12, 2019 1:53 pm, edited 1 time in total.

User avatar
epilectrolytics
Donor
Donor
Posts: 498
Joined: Thu Oct 04, 2018 6:06 am

Re: PhotonGI cache

Post by epilectrolytics » Tue Feb 12, 2019 12:51 pm

The last version is a large step forward.
Everything is faster and smoother and easier.
I wonder if the initial photoncount parameter could be dropped and instead photons are continuously shot until maxsize parameter is reached?
And in debug mode the rendering could be stopped after cache is complete and the image drawn.

I'm seeing the same light leak issues as lacilaci in my tests.
B.Y.O.B. wrote:
Tue Feb 12, 2019 12:07 am
I have updated the Blender addon with the latest property changes.
Both new "usagethreshold" parameters are not exposed?
MBPro 15" 16GB i7-4850HQ GT750M, MacOS 10.13.6 & Win10Pro PC 16GB Ryzen 2700X, 2 x RTX 2070

User avatar
lacilaci
Donor
Donor
Posts: 1381
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Tue Feb 12, 2019 1:36 pm

epilectrolytics wrote:
Tue Feb 12, 2019 12:51 pm
The last version is a large step forward.
Everything is faster and smoother and easier.
I wonder if the initial photoncount parameter could be dropped and instead photons are continuously shot until maxsize parameter is reached?
And in debug mode the rendering could be stopped after cache is complete and the image drawn.

I'm seeing the same light leak issues as lacilaci in my tests.
B.Y.O.B. wrote:
Tue Feb 12, 2019 12:07 am
I have updated the Blender addon with the latest property changes.
Both new "usagethreshold" parameters are not exposed?
Probably depth could be default 8 and hidden (4 is too low for interior spaces and it shows a lot, same goes for diffuse and glossy rays for pathtracing)
I wonder if lookup radius could be defaulted as well. From all my tests 20-30 cm would be enough but we now have those lightleaks.

As for photon count and cache size, I've noticed that with fixed lookup (for example 30cm) but higher values for photon count and cache size can give more precise looking result. Maybe there could be a way to estimate some sort of accuracy and have luxcore keep shooting and resizing cache until certain level of accuracy is reached? Is that technically possible?

User avatar
lacilaci
Donor
Donor
Posts: 1381
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Tue Feb 12, 2019 2:26 pm

comparison between
60mil cache + 60mil size - 10cm lookup
60mil_cache.jpg
10mil cache + 10mil size - 10cm lookup
10mil_cache.jpg

User avatar
Dade
Developer
Developer
Posts: 2820
Joined: Mon Dec 04, 2017 8:36 pm

Re: PhotonGI cache

Post by Dade » Tue Feb 12, 2019 5:17 pm

I tried to apply the same idea to caustic cache but while it was a LOT faster, the quality of the output was too low to be really usable so I leave caustic cache as it is now. I will do only another test trying to filtering the caustic cache entries to smooth them and avoid the "point-iness" looking it can have sometime.
Support LuxCoreRender project with salts and bounties

User avatar
lacilaci
Donor
Donor
Posts: 1381
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Tue Feb 12, 2019 5:43 pm

Dade wrote:
Tue Feb 12, 2019 5:17 pm
I tried to apply the same idea to caustic cache but while it was a LOT faster, the quality of the output was too low to be really usable so I leave caustic cache as it is now. I will do only another test trying to filtering the caustic cache entries to smooth them and avoid the "point-iness" looking it can have sometime.
shame, I was hoping that this could produce reflective caustics

User avatar
Dade
Developer
Developer
Posts: 2820
Joined: Mon Dec 04, 2017 8:36 pm

Re: PhotonGI cache

Post by Dade » Tue Feb 12, 2019 11:35 pm

lacilaci wrote:
Tue Feb 12, 2019 11:21 am
Dade wrote:
Tue Feb 12, 2019 11:14 am
lacilaci wrote:
Tue Feb 12, 2019 10:18 am
here's how that bleeding looks. I'm getting this everywhere (probably related to glossy reflection ?)
My Blender 2.79b says the scene has been written with 2.80 and crashes :?:
I'm sorry, wrong file...

here:
00_concept_lux.zip
I have found a bug: Metal material was claiming to be "GLOSSY | REFLECT" and was misidentified as Glossy material by PhotonGI. This could lead to some problem.

The rendering now looks ok to me:

render.jpg

Your cache is also starving for more photons. Follow 3 rendering with 5, 25 and 250 millions of photons traced:

5m.jpg
25m.jpg
250m.jpg

Tracing 250 millions of photons requires 100secs so it is not a totally insane setting and it gives you the idea of how much light you are missing in your rendering (and 25mils is already a lot better than your 5mils).
Support LuxCoreRender project with salts and bounties

User avatar
lacilaci
Donor
Donor
Posts: 1381
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Wed Feb 13, 2019 5:31 am

Dade wrote:
Tue Feb 12, 2019 11:35 pm
lacilaci wrote:
Tue Feb 12, 2019 11:21 am
Dade wrote:
Tue Feb 12, 2019 11:14 am


My Blender 2.79b says the scene has been written with 2.80 and crashes :?:
I'm sorry, wrong file...

here:
00_concept_lux.zip
I have found a bug: Metal material was claiming to be "GLOSSY | REFLECT" and was misidentified as Glossy material by PhotonGI. This could lead to some problem.

The rendering now looks ok to me:


render.jpg


Your cache is also starving for more photons. Follow 3 rendering with 5, 25 and 250 millions of photons traced:


5m.jpg
25m.jpg
250m.jpg


Tracing 250 millions of photons requires 100secs so it is not a totally insane setting and it gives you the idea of how much light you are missing in your rendering (and 25mils is already a lot better than your 5mils).
I'll try the fix.

Also, regarding the lighting difference, I don't think it's about photons traced although that makes difference too.
I've seen pretty big differences in open space where it shouldn't be a problem with the number of photons.. Will post more tests

User avatar
lacilaci
Donor
Donor
Posts: 1381
Joined: Fri May 04, 2018 5:16 am

Re: PhotonGI cache

Post by lacilaci » Wed Feb 13, 2019 8:41 am

so, no... Amount of photons have some effect on brightness, but not even close to pathtracing.

Higher photon count helps with leaks (which are still present near corners!) but have only limited effect on brightness of GI

here's 250 000 000 photons with 250 000 000 cache size VS pathtracing. Notice that even top right corner which is very easily accessible for GI is still darker.
PTvsPGI.jpg
and here's only 250 000 photons and cache size! Note that between 250mil and 250 thousand cache there is only a difference in color balance, but brightness is the same :!:
250kilo_photons.jpg
So my conclusion so far is, we don't need hundred of millions photons, we need more brightness, and while we can use more photons and larger lookup radius to combat leaks, there has to be something wrong cause they only appear near corners and are always brighter than samples further from the corner (Or maybe, they have correct brightness but the ones further from corners are wrong and too dark!)!

Another problem, I don't know what are the latest changes to materials (I've read on github material list fix so I assumed now all materials work)
But, now even my matte wall is not able to gather GI samples from sky (note that this scene contains only sky) EDIT: it actually does gather photon gi but it is very very dark, in older builds it wasn't that bad.
Screenshot from 2019-02-13 09-47-23.png

User avatar
Dade
Developer
Developer
Posts: 2820
Joined: Mon Dec 04, 2017 8:36 pm

Re: PhotonGI cache

Post by Dade » Wed Feb 13, 2019 11:19 am

lacilaci wrote:
Wed Feb 13, 2019 8:41 am
Another problem, I don't know what are the latest changes to materials (I've read on github material list fix so I assumed now all materials work)
Nothing has changed, cache entries are still created only on matte and glossy material (due to a now fixed bug, before they were created on metal material too).
lacilaci wrote:
Wed Feb 13, 2019 8:41 am
But, now even my matte wall is not able to gather GI samples from sky (note that this scene contains only sky) EDIT: it actually does gather photon gi but it is very very dark, in older builds it wasn't that bad.
Screenshot from 2019-02-13 09-47-23.png
This was the result of a typo, causing the lookup radius parameter being used instead of the lookup normal threshold parameter. This was stopping PhotonGI from creating entries on some or even all surfaces. I should have fixed this problem.
Support LuxCoreRender project with salts and bounties

Post Reply