Announcement
Collapse
No announcement yet.
Noise threshold per object or material?
Collapse
X
-
well the old rQmc sampler had a different threshold. Its changed now, you don't need to have it that low.Dmitry Vinnik
Silhouette Images Inc.
ShowReel:
https://www.youtube.com/watch?v=qxSJlvSwAhA
https://www.linkedin.com/in/dmitry-v...-identity-name
Comment
-
ok, so any other general strategies that I could try?
Despite years of trying, I've generally never been happy with using irradiance map as a GI method, so I'm mostly interested in Brute Force & Light Cache
I did have one project where I really struggled to get acceptable images, and I used pretty much every render element I could to try and identify where the noise was coming from, so I understand what to do up to that point, I just dont know how to use that info in a productive way.
Comment
-
I think it all depends on the render-time vs user time spent on the image. For example if you have a formula that works, but you know it will take a day to render an image, its better to do that then spend a day trying to speed it up and at the end of the day get something crappy. After many, many years of doing pretty much same thing, I ended up making the render longer/better, and letting the computer do the job for me.
That's not always the acceptable scenario though, but you typically wouldn't render 4k animations and for smaller frames its quite acceptable time.
Im going to make a small doc about the dmc sampler in a bit, perhaps it will help.Dmitry Vinnik
Silhouette Images Inc.
ShowReel:
https://www.youtube.com/watch?v=qxSJlvSwAhA
https://www.linkedin.com/in/dmitry-v...-identity-name
Comment
-
Hey Peter
I work more or less exactly as you do, and for the same reasons. I'd rather waste my computer's time than my own, and I gave up on trying to get consistently good quality renders with IM/LC. I see it done a lot, but it just isn't the thing for me I guess.
Anyway, the noise issue is areas like that was something I found no good solution for, other than (1) increasing light samples quite a bit, especially for dome lights. I generally set Vraylight shadows to 60-80 as a base now, but dome lights I set to 100-200. That seemed to help a fair bit. (2) is just lowering the sampler threshold as you have found. I typically render at .01 to maybe .008 (my renders are around the size of yours, 4k-8K) but some things just don't clean up until you get down to .005 or even lower.
Typically I can live with the noise though, as it feels more like grain and blends better with photos (even high end digitals produce some noise, so perfectly smooth looks fake even when composited with shots from a high end Hasselblad etc.)
There may well be better options out there, but I'm more or less in the same boat as you anyway
b
Comment
-
Thanks for your thoughts Brett. I'll definitely play about with vraylight subdivs. When used with universal settings (or similar), does changing the subdivs of a light from 8 to 60-80 make much difference? I did a series of tests with dome light subdivs a while back and actually found that really high subdivs seemed to do weird things to the lighting... but i'll need to re-test to confirm that
I certainly don't mind a bit of noise either, and at 4k 0.01 is usually perfectly acceptable for me. I guess the reason I started this thread though is that at 4k, for the most part a threshold of 0.02 would actually be fine, and for that reason it would be cool to be able to change the threshold for certain objects or materials.
My finished work is generally pretty noisy at 4k actually, mostly I'm happy enough that it looks fine when resized down to screen res, and don't care so much about what the clients receive! (ssssh dont tell)(they are usually happy enough anyway). But I can see that in your and other people's work the image quality at high res is more important.
Comment
-
I was having some trouble recently getting some soft contact shadows clean with a product on a white background. In that case bumping up the dome light and other lights made a more appreciable difference than in most cases - I think it really depends, but on what I couldn't say I also ended up dropping down to .008 and the two together gave me a cleaner result, but the render time was longer, but it was going to be long anyway.
For me noise is okay in most cases too, I just have to watch it doesn't get too crazy because a lot of the work gets blown up to billboards and bus-shelter posters etc, so it can become an issue for that kind of stuff. However - you are right in that it probably is more of an issue for me than anyone else. Clients often don't notice and are often happy with billboards they blow up from postage-stamp crops from shitty digital stock
/b
Comment
-
Originally posted by Morbid Angel View PostIm going to make a small doc about the dmc sampler in a bit, perhaps it will help.
I'm also tired of working with IRMaps, but they do render quicker. I believe Brute Force is a much better solution, if I can just fully understand the dmc sampler and AA then maybe I can get my renders a bit quicker with acceptable quality.
I tried your settings you mentioned in another thread and I got super clean results, now I just need to understand why I got clean results so I can tweak it a bit to bring down the times.
For example I'm now dealing with YET ANOTHER problematic scene even with IRMap I'm getting times of 6 HOURS for 1 frame at 720p That is WAY TOO MUCH! Its fine if at 4k it renders 6 hours, but at 720p I would expect it to be not more than say 30 minutes.
It's a mall with curved passage. I have an override material of 200,200,200 with a white reflection and fresnel ticked, glossiness at 0.85 subdivs at 16. I excluded the shopfront glass from the override. So, its just the passage with nothing in it, and the EMPTY shops on the sides with NO lighting yet in the shops. The passage is being lit by ONLY 3 cove lights running the lenght of the passage. It's noisy like you cant believe and even like that it wants to render 4 hours. If I attampt to clean the noise the render times goes up like you wont believe!
Yes I know it will be better once the shops have their own lighting spilling into the passage, but I want to at least know its possible to get clean renders as is with acceptable quality and times.
My cove lighting is VRayLightMTL and has colour 255,230,175 and multiplier 3, compensate camera exposure ticked, direct lighting ticked with subdivs 64 (tried 128 also)Last edited by Morne; 27-04-2012, 01:26 AM.Kind Regards,
Morne
Comment
-
-
Originally posted by CA Portugal View PostHi there,
I would love to have such control over each object.
Peter, a question about BF. What DMC you normally use in your finals? 100 like spot3d suggests?
Cheers
Comment
-
Yes, but the big problem with Universal Setting (US) vs IM+LC is the sub-pixel mapping and the rendering time.
If you want to have the same result, you must un-check sub-pixel mapping with US. But rendering time go up x3 or x4 !
Without, rendering time go down a lot, but with much more noise and the dark result. The differences are near the bright area and/or light sources and/or glossy surface.
So, I think Peter an all other our friends, usually, use US + sub-pixel mapping ON.
Really, now I start to reconsidering the old IM+LC or BF+LC but not in US. But yes, a lot of waste time in tweaking (((Last edited by cecofuli; 03-05-2012, 12:43 AM.
Comment
-
For me, there are so many situations where having sub pixel mapping on results in an image that doesn't look as good and as real as having it off. For example, trees against a bright sky: there is a huge difference and having sub pixel mapping on results in a very CG look IMO. So I pretty much do all my renders with it off. (can't think of a project in the last 3 years where I didn't)
Comment
Comment