Hi all,
I am rendering a scene with some neurons and the scene is flickering for some reason. You can see what I mean here: https://youtu.be/AA6ozUMxfUo
I have read through past forum posts (coherent ratio is set to zero, I toggled on/off AI denoising, I even tried rendering direct lighting in stead of path tracing, I also restarted and removed and redid the HDRI). Could anyone help troubleshoot? Here is the project file: https://drive.google.com/file/d/1fiQAVY ... drive_link
Thank you,
Amy
Flickering
Moderators: ChrisHekman, aoktar
This looks like a classic Ray Epsilon issue; this scene is absolutely HUGE. That central yellow neuron trunk alone is is over 6,000,000 cm (2,000 feet) wide. I eliminated the flicker by cranking the Ray Epsilon up to 0.5. (Ray Epsilon is in the top group of attributes in your Kernel settings.)
The Ray Epsilon value essentially controls how close to a surface a ray has to get to count as hitting/intersecting with it. With a giant scene scale and a tiny ray epsilon value you are asking Octane to sample space with a much higher precision than it is designed for. It just doesn't have the floating point accuracy to illuminate the surface reliably.
Personally, I'd be more much more comfortable scaling the scene down to numbers I can reasonably juggle in my head, and where light and surface interactions behave more intuitively. I try to work with a scene scale somewhere between tens of meters down to a few cm or so most of the time.
p.s.
By, "doesn't have the floating point accuracy to illuminate the surface reliably." I mean that the surface can intermittently cast its shadow in front of itself instead of behind, or in funky moire patterns. You can also have problems with polygons that are nearly coplanar popping in and out because the spatial resolution isn't fine enough to determine which surface is in front and which is in back (also called z-fighting, plane fighting, stitching, etc.)
Broadly speaking, the larger the scale of the full set of elements that need to be in-frame at any given moment, the larger the RE has to be to keep the orders of magnitude between world space and surface precision within a reasonable range. The default value of 0.0001, or 1/10,000 RE works fine for many tens of meters down to a few cm, but above or below those scales, you may need to tweak the RE up or down, or consider scaling the project geometry to render properly within that range. This can be tricky if you are trying to combine micro and macro elements in a single image, like seeing an ultra-closeup of an ant with a large landscape also visible in the background. In those cases, you may want to render foreground and background separately then composite because they can require very different REs to render reliably.
The Ray Epsilon value essentially controls how close to a surface a ray has to get to count as hitting/intersecting with it. With a giant scene scale and a tiny ray epsilon value you are asking Octane to sample space with a much higher precision than it is designed for. It just doesn't have the floating point accuracy to illuminate the surface reliably.
Personally, I'd be more much more comfortable scaling the scene down to numbers I can reasonably juggle in my head, and where light and surface interactions behave more intuitively. I try to work with a scene scale somewhere between tens of meters down to a few cm or so most of the time.
p.s.
By, "doesn't have the floating point accuracy to illuminate the surface reliably." I mean that the surface can intermittently cast its shadow in front of itself instead of behind, or in funky moire patterns. You can also have problems with polygons that are nearly coplanar popping in and out because the spatial resolution isn't fine enough to determine which surface is in front and which is in back (also called z-fighting, plane fighting, stitching, etc.)
Broadly speaking, the larger the scale of the full set of elements that need to be in-frame at any given moment, the larger the RE has to be to keep the orders of magnitude between world space and surface precision within a reasonable range. The default value of 0.0001, or 1/10,000 RE works fine for many tens of meters down to a few cm, but above or below those scales, you may need to tweak the RE up or down, or consider scaling the project geometry to render properly within that range. This can be tricky if you are trying to combine micro and macro elements in a single image, like seeing an ultra-closeup of an ant with a large landscape also visible in the background. In those cases, you may want to render foreground and background separately then composite because they can require very different REs to render reliably.
Animation Technical Director - Washington DC