Lights with radius 0 produce a dark ring in EEVEE #82437

New Issue

Cycles · 2020-11-05T18:30:53+01:00

Cycles commented

2020-11-05 18:30:53 +01:00

System Information
Operating system: Microsoft Windows 10 Pro
Graphics card: Nvidia GeForce GTX 1050
Processor: Intel(R) Core(TM) i5-2500 CPU @ 3.30GHz, 3301 Mhz, 4 Core(s), 4 Logical Processor(s)

Blender Version
Broken: blender-2.91.0-02677ec4e0fa-windows64

Short description of error
In EEVEE, objects with certain roughness values have a ring inside the specular highlight in which there is no light, when the light source has a radius of 0. Changing the radius to have a higher value makes the ring go away. Setting the value to near zero values such as 0.002 will still have the ring, but smaller. Both distributions GGX and Multiscatter GGX have the problem. Only the lights with a radius parameter have the ring (point and spot lights). Area lights use a size parameter and behave differently, so they do not have this issue. Sun lights have an angle parameter instead, and the fact that they are treated as having an infinite distance may also play a role.

Exact steps for others to reproduce the error
specular_ring.blend
Open blend file
Set mode to rendered view
Notice a dark ring around the specular highlight
Change radius to value over 0.005
The ring goes away

**System Information** Operating system: Microsoft Windows 10 Pro Graphics card: Nvidia GeForce GTX 1050 Processor: Intel(R) Core(TM) i5-2500 CPU @ 3.30GHz, 3301 Mhz, 4 Core(s), 4 Logical Processor(s) **Blender Version** Broken: blender-2.91.0-02677ec4e0fa-windows64 **Short description of error** In EEVEE, objects with certain roughness values have a ring inside the specular highlight in which there is no light, when the light source has a radius of 0. Changing the radius to have a higher value makes the ring go away. Setting the value to near zero values such as 0.002 will still have the ring, but smaller. Both distributions GGX and Multiscatter GGX have the problem. Only the lights with a radius parameter have the ring (point and spot lights). Area lights use a size parameter and behave differently, so they do not have this issue. Sun lights have an angle parameter instead, and the fact that they are treated as having an infinite distance may also play a role. ![specular_ring.png](https://archive.blender.org/developer/F9210819/specular_ring.png) **Exact steps for others to reproduce the error** [specular_ring.blend](https://archive.blender.org/developer/F9210817/specular_ring.blend) Open blend file Set mode to rendered view Notice a dark ring around the specular highlight Change radius to value over 0.005 The ring goes away

Cycles commented

2020-11-05 18:30:53 +01:00

Added subscriber: @Cycles

Cycles changed title from ~~Bright lights with radius 0 produce a dark ring in EEVEE~~ to Lights with radius 0 produce a dark ring in EEVEE

2020-11-06 01:37:40 +01:00

Philipp Oeser commented

2020-11-06 12:22:29 +01:00

Changed status from 'Needs Triage' to: 'Confirmed'

Cycles commented

2020-11-12 18:55:26 +01:00

Trying to find a consistent pattern to when this happens is proving to be difficult. Sometimes, a rough specular highlight has no ring and looks fine. Other times, the highlight seems to disappear entirely, especially when the light is more distant. It looks a lot worse on metallic objects but increase the light's specular value makes the problem noticeable even on non metallic objects. Sometimes, it appears as a clipped highlight rather than a ring. It happens very randomly and I am not sure why.

Jishan Singh commented

2020-11-14 23:26:41 +01:00

Added subscriber: @ringmaker

Jishan Singh commented

2020-11-14 23:26:41 +01:00

i m new developer here. here's what i found
error is caused in ltc_lib.glsl :- ltc_evaluate_disk
cause: due to small value of radius roots evaluated of cubic equation are very close to 0
three roots a,b,c (ascending order) a and b are too close to 0 and a is being taken as 0 while we calculate L = a/b it evaluates to 0 causing error

i m new developer here. here's what i found error is caused in ltc_lib.glsl :- ltc_evaluate_disk cause: due to small value of radius roots evaluated of cubic equation are very close to 0 three roots a,b,c (ascending order) a and b are too close to 0 and a is being taken as 0 while we calculate L = a/b it evaluates to 0 causing error

Cycles commented

2020-11-18 00:59:03 +01:00

Quoted Text

In #82437#1055003, @ringmaker wrote:
i m new developer here. here's what i found
error is caused in ltc_lib.glsl :- ltc_evaluate_disk
cause: due to small value of radius roots evaluated of cubic equation are very close to 0
three roots a,b,c (ascending order) a and b are too close to 0 and a is being taken as 0 while we calculate L = a/b it evaluates to 0 causing error

So this is a precision error? I wonder what a workaround case for this would look like. Maybe it can clamp the value just above zero to avoid the error? I don't know how this would be implemented or how practical it would be, but maybe something like this would work?

> Quoted Text > In #82437#1055003, @ringmaker wrote: > i m new developer here. here's what i found > error is caused in ltc_lib.glsl :- ltc_evaluate_disk > cause: due to small value of radius roots evaluated of cubic equation are very close to 0 > three roots a,b,c (ascending order) a and b are too close to 0 and a is being taken as 0 while we calculate L = a/b it evaluates to 0 causing error So this is a precision error? I wonder what a workaround case for this would look like. Maybe it can clamp the value just above zero to avoid the error? I don't know how this would be implemented or how practical it would be, but maybe something like this would work?

Clément Foucault commented

2021-01-26 13:07:26 +01:00

Added subscriber: @fclem

Clément Foucault commented

2021-01-26 13:07:26 +01:00

This issue has been reported before. Unfortunately we cannot clamp since 0 is a legitimate value too. We already clamp roughness and light radius to a very small value, but this artifact depends on viewing condition so it is quite not possible to fix without improving the actual LTC algorithm. Which is way above my knowledge.

If someone wants to take a look at it feel free to do it.

This issue has been reported before. Unfortunately we cannot clamp since 0 is a legitimate value too. We already clamp roughness and light radius to a very small value, but this artifact depends on viewing condition so it is quite not possible to fix without improving the actual LTC algorithm. Which is way above my knowledge. If someone wants to take a look at it feel free to do it.

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Lights with radius 0 produce a dark ring in EEVEE #82437