SkIndigo 1.1.16 and the fireflies
SkIndigo 1.1.16 and the fireflies
Sorry guys to report this again, but it's starting to get me a little angry..
With complex scenes (and, you know, architectural projects ARE complex scenes) I always have huge amount of fireflies, which make the render noisy and a little bothering. Obviously, longer render times won't lead to better results. Is it something which can be fixed or something? I really hope so, 'cos I really (professionally) cannot count on a renderer that makes such a bothering noise randomly in the final image.
I really don't understand why, cos I've already rendered other scenes of the same model, and they were (are) just fine, really good indeed. Whites are not too white, there is no glass unless necessary...
Tips? Suggestions? Any strategies to avoid this effect?
Thank you very much!
Oh, guys: don't get me wrong. I'm really grateful to you for all the work you've done. I'm only reporting this issue. Perhaps there's something we can do to make it better, that's all.
With complex scenes (and, you know, architectural projects ARE complex scenes) I always have huge amount of fireflies, which make the render noisy and a little bothering. Obviously, longer render times won't lead to better results. Is it something which can be fixed or something? I really hope so, 'cos I really (professionally) cannot count on a renderer that makes such a bothering noise randomly in the final image.
I really don't understand why, cos I've already rendered other scenes of the same model, and they were (are) just fine, really good indeed. Whites are not too white, there is no glass unless necessary...
Tips? Suggestions? Any strategies to avoid this effect?
Thank you very much!
Oh, guys: don't get me wrong. I'm really grateful to you for all the work you've done. I'm only reporting this issue. Perhaps there's something we can do to make it better, that's all.
- Attachments
-
- This after 21 hours of calculation...
- fireflies.jpg (453.99 KiB) Viewed 18319 times
A few hints Pibuz, I'm sure you know some already but why not remind them ?
- use a larger super sample value, in the range 3-5. Yes, that's big, but hopefully the wider the frame buffer, the more isolated nuked pixels will be. That should allow the render to overally reach a satisfying convergence while helping to nuke the bug
- make sure specular materials (specially transparent ones) are not intersecting with a medium-less ones (phong, diffuse etc), 0,2mm is about the minimal safe value between transparent geometries (down to 0,1 mm) and any others. Of course mediums can intersect other mediums as expected.
- In last resort give a shot to plain MLT ?
- use a larger super sample value, in the range 3-5. Yes, that's big, but hopefully the wider the frame buffer, the more isolated nuked pixels will be. That should allow the render to overally reach a satisfying convergence while helping to nuke the bug

- make sure specular materials (specially transparent ones) are not intersecting with a medium-less ones (phong, diffuse etc), 0,2mm is about the minimal safe value between transparent geometries (down to 0,1 mm) and any others. Of course mediums can intersect other mediums as expected.
- In last resort give a shot to plain MLT ?
Hi guys thank for the support!
FUSED: I'm running Indigo 1.1.16, with skindigo 1.1.16. The render you see is calculated with BiDir PT.
CTZn: thanks for the hints mate. Now I'm trying to render with plain MLT to see the result. If the damn fireflies still show up I'll try modifying the super sample value: does a bigger number affect the "unbiasing" quality?
Specular mats don't intersect each other at all. In some cases I have TG geometry (window) colliding with a phong mat (metallic frame), but I hope this isn't the actual problem. I'd have serious problems detaching every window profile from its frame...
FUSED: I'm running Indigo 1.1.16, with skindigo 1.1.16. The render you see is calculated with BiDir PT.
CTZn: thanks for the hints mate. Now I'm trying to render with plain MLT to see the result. If the damn fireflies still show up I'll try modifying the super sample value: does a bigger number affect the "unbiasing" quality?
Specular mats don't intersect each other at all. In some cases I have TG geometry (window) colliding with a phong mat (metallic frame), but I hope this isn't the actual problem. I'd have serious problems detaching every window profile from its frame...
- PureSpider
- Posts: 1459
- Joined: Tue Apr 08, 2008 9:37 am
- Location: Karlsruhe, BW, Germany
- Contact:
- kellpossible
- Posts: 97
- Joined: Thu Nov 27, 2008 5:46 pm
- Location: Australia
I posted a similar problem earlier, and switching to 1.1.18 fixed most of the problem, but I'll give super sampling a shot too. Just wondering, but could you achieve a similar effect to super sampling by scaling down the image after it has been rendered or is the process a lot more complicated than that.
(I'm sure that it would also depend on what image editor you used to do the scaling!)
(I'm sure that it would also depend on what image editor you used to do the scaling!)
Super sampling in Indigo is just a resizing of the frame buffer with a choosen filter, amongst the 3 available. So it can very well be applied in post by yourself kellpossible 
About filters: the splat_filter should not be mn_cubic with default settings since it is putting the emphasis on nuked pixels. The splat_filter is the one used to apply pixels onto the native frame buffer, while the downsize_filter is the one that does the reduction job.
If the filters are not exposed in your exporter then you certainly want to set super_sample_factor to 1 and reduce in post.

About filters: the splat_filter should not be mn_cubic with default settings since it is putting the emphasis on nuked pixels. The splat_filter is the one used to apply pixels onto the native frame buffer, while the downsize_filter is the one that does the reduction job.
If the filters are not exposed in your exporter then you certainly want to set super_sample_factor to 1 and reduce in post.
Who is online
Users browsing this forum: No registered users and 42 guests