PhotonGI cache

[lacilaci]

Another issue,

I started testing pgi on my office scene and I'm seeing a lot of reflections being cut short...
I've noticed this happening in every scene.

Only way to fix this seems to be using very high Usage threshold scale.
For example, lookup radius 10cm and usage threshold scale of 6 was used in the rendering.
To fix the reflection problem I had to use usage threshold scale of 16 (I only tried the fix on crop render so I don't know what's the performance penalty of using higher usage threshold)

[Asticles]

What does this threshold do?

I feel we will need a good troubleshooting guide to use this feature.

[Racleborg]

The scene (attached) crashes every time with PGI on, and any scene I create with a MIX material like this, also crashes.


Thank you

[Dade]

(reseted scale, shading, collapsed modifiers and turned ngons to triangles etc...) they just don't gather photons

It looks like another NaNs problem, I will check.

Surprisingly, it looks like the problem is in the geometry. First of all, it renders fine on the CPU while the problem is visible in OpenCL. This is because the (mysterious) reduced precision of the OpenCL code. If you check the geometry, you will see the object has 1 back face overlapped with the wall face (and there is a "Z fight" even inside Blender shaded view port). Than the front face is very near the back face. Deleting the back face and making an hole in wall face (so it is not near the front face) should fix the problem.

[Dade]

So, why does the default normal angle kill so much of the light? 10° seems very low

90deg normal angle:
1min13_90deg.jpg
10deg normal angle:
1min13_10deg.jpg

10degree normal angle kills a lot of GI light and also seems to be darker than pathtracing only..

It may be exactly the opposite: floor will start to "borrow" light form perpendicular walls using a 90 deg angle so the image will start to gain too much light (instead of loosing them when using a 10deg angle).

This seems also suggested by the reference rendering with plain path tracing: https://github.com/LuxCoreRender/LuxCor ... erence.png

It is darker, not brighter.

[Dade]

I started testing pgi on my office scene and I'm seeing a lot of reflections being cut short...
I've noticed this happening in every scene.

Only way to fix this seems to be using very high Usage threshold scale.
For example, lookup radius 10cm and usage threshold scale of 6 was used in the rendering.
To fix the reflection problem I had to use usage threshold scale of 16 (I only tried the fix on crop render so I don't know what's the performance penalty of using higher usage threshold)

Just leave the default parameters and mark the material of the floor as "not cached". It will fix the problem with a barely noticeable impact on rendering time.

Have you changed the default glossiness threshold ? I ask because looking at the floor, I'm quite sure it has a roughness lower than 0.2 (i.e. it is nearly a mirror).

[lacilaci]

Yes I did use glossines threshold at 0.0
Probably not a good idea.

[Dade]

The scene (attached) crashes every time with PGI on, and any scene I create with a MIX material like this, also crashes.

This is somewhat expected, your scene includes a single cube and the material roughness is under 0.2 so it is marked as "cache disabled". So you entire scene doesn't include a single surface where cache is enabled and PhotonGI is useless.

I have modified the code so it will print a big "WARNING" for this case and will proceed rendering with normal path tracing.

[Dade]

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)
test.jpg

I'm getting exactly the same rendering (caustic aside) out of PathOCL+PhotonGI:

and PathOCL:

I'm using this light source:

Code: Select all

scene.lights.distl.type = distant
scene.lights.distl.color = 1.0 1.0 1.0
scene.lights.distl.gain = 10 10 10
scene.lights.distl.direction = 1 1 -1

Are you using a different type of light source ?

[B.Y.O.B.]

I assume epilectrolytics is using a constantinfinite light.