Page 49 of 109
Re: PhotonGI cache
Posted: Thu Feb 21, 2019 12:55 pm
by epilectrolytics
alpistinho wrote: ↑Wed Feb 20, 2019 8:03 pm
epilectrolytics wrote: ↑Wed Feb 20, 2019 7:56 pm
Dade wrote: ↑Wed Feb 20, 2019 4:53 pm
I have introduced a new stage in PhotonGI caustic cache pre-processing. It merges near photons coming from the same directions...
Please avoid compromising core features like caustic sharpness or the result will be unusable for serious work.
The goal should be correct caustics plus SDS paths, speed is of lesser concern.
There is a parameter for this merging, so I guess you can just disable it if you want
Thanks goodness you're right!
Very embarrassing I didn't get it when it's right in the title
I was just so scared when I tried it on my test scene and the caustics turned blurry!
Background light caustics still don't work right though.
Editing B.Y.O.B.'s test file by setting the marge.radiusscale to 0.01 I could get decent sharpness with still some speed gain in OCL.
Re: PhotonGI cache
Posted: Thu Feb 21, 2019 4:57 pm
by Dade
lacilaci wrote: ↑Thu Feb 21, 2019 11:36 am
so apparently, bump mapping is still killing cache for me.
EDIT: I'm actually not sure if it's bump or something else, but I couldn't figure out what is going on... everything is mostly black
The smell like floating point NaNs (i.e. Not a Number, like the result of a division by zero, etc.). I will check.
Re: PhotonGI cache
Posted: Thu Feb 21, 2019 6:30 pm
by epilectrolytics
Another visualisation of the problem the caustic cache has with infinite background light:
Took B.Y.O.B.'s dice file and replaced the area light with white background color:
(latest builds including photon merge, from Blender)
Re: PhotonGI cache
Posted: Thu Feb 21, 2019 10:24 pm
by Dade
Dade wrote: ↑Thu Feb 21, 2019 11:20 am
marcatore wrote: ↑Thu Feb 21, 2019 11:06 am
Something to suggest?
It is this bug:
normal.jpg
See the border on the glossy material ? I have the felling it is due to cached paths not including the light emission, I'm looking into the problem.
I should have fixed the problem:
Re: PhotonGI cache
Posted: Thu Feb 21, 2019 10:39 pm
by Dade
Dade wrote: ↑Thu Feb 21, 2019 4:57 pm
lacilaci wrote: ↑Thu Feb 21, 2019 11:36 am
so apparently, bump mapping is still killing cache for me.
EDIT: I'm actually not sure if it's bump or something else, but I couldn't figure out what is going on... everything is mostly black
The smell like floating point NaNs (i.e. Not a Number, like the result of a division by zero, etc.). I will check.
I can confirm that your scene is full of:
There must be a material doing something nasty, I'm looking where it is.
Re: PhotonGI cache
Posted: Thu Feb 21, 2019 11:41 pm
by Dade
Dade wrote: ↑Thu Feb 21, 2019 10:39 pm
I can confirm that your scene is full of:
There must be a material doing something nasty, I'm looking where it is.
I fixed the problem: it was glossy coating material to spit out NaNs. This is a debug output now:
and this is the denoised output after 32 samples:
Re: PhotonGI cache
Posted: Fri Feb 22, 2019 6:11 am
by lacilaci
Oh yes, things are looking much better now.
I did notice though that with some materials I have to raise normal angle quite high ~60 degree, to get gi caching working on them (most likely due to bump mapping)
Re: PhotonGI cache
Posted: Fri Feb 22, 2019 7:55 am
by lacilaci
seems that some recent changes made light leaking worse... nothing helps
Re: PhotonGI cache
Posted: Fri Feb 22, 2019 11:21 am
by Dade
lacilaci wrote: ↑Fri Feb 22, 2019 6:11 am
I did notice though that with some materials I have to raise normal angle quite high ~60 degree, to get gi caching working on them (most likely due to bump mapping)
It is not possible because geometry normal is now used. I can not add anything more without a test scene.
Re: PhotonGI cache
Posted: Fri Feb 22, 2019 11:22 am
by Dade
lacilaci wrote: ↑Fri Feb 22, 2019 7:55 am
seems that some recent changes made light leaking worse... nothing helps
I can not add anything more without a test scene.