I’m posting this here because i am not at all sure it’s not me doing something wrong, rather than VRay having an issue.
What I (and some other old-time user as well) am finding out is that the new DMC sampler is a lot harder to work with, to get clean results.
There are a few cases where it’s nigh impossible to clean an image, for instance when using an LC/IRmap GI solution coupled with a web IES light, and there are very fine details in the scene.
Every trick that used to work with the rQMC seems to fail, producing little to no change (upping samples, lowering thresholds, brightening/darkening the image to accommodate for an easier time on the sampler,etc.).
With rQMC, if i wanted a super-clean image, I’d only have to lower thresholds, raise samples, and it would every time give me the expected result, notwithstanding the rendertimes, which when sure of the final quality can be secondary.
Now the question is fairly simple: is there a parameter which has more influence than it used to have with rQMC, some I have badly overlooked? Or is the DMC sampler still being worked on and improved, and I’m to expect it to get better with the updates?
I am with Lele there. Impact of the settings seems to have changed a lot and it seems a bit cumbersome to find good settings for certain scenes. Detailed Chrome causes quite sever Aliasing even if Reflections are kept <1 and it’s pretty hard to get rid of.
If you have a scene that is giving you trouble, would be best to post it here or email it to me at vlado@chaosgroup.com to see if there is anything wrong, since the reason may not be in the DMC sampler at all. In any case, the old sampler cannot be returned back. There are some improvements to the new one for SP2 though.
I found new sampler to be more random, than older. But, produces better antialiasing, with small perfomance hit. Maybe now we should increase AA settings ?
Will send you the scene in no time, just trying to make it readable (for the issues) before sending it over.
Thankyou very much for the open and exhaustive answer.
I run into problems with noise, too.
doesn’t care wich method I take, there’s always noise.
I think the DMC is the problem.
vray will become like maxwell more and more.
my settings: AA: ADMC 1/16, IRR/DE/LC, DMC: 0.85/0.002/16/1
images are rendering 22h@3500px, that’s too long.
with such a low noise threshold you’re likely forcing the DMC AA sampler towards max subdivs (ie. 256 samples taken for each pixel, regardless of what’s actually in the pixel) as soon as an area light or a glossy effect gets calculated.
That threshold will also slow down the IRMap sample gathering.
My suggestion is to raise that noise threshold to something like 0.005-0.0075, and at most raise max subdivs.
If you want a clearer picture of how the sampling is going in your particular shot, try and use the newest addition to the matrix exposimeter.
It’ll tell you what’s the average sampling rate for your image (in your case, ideally 8, as you have min 1 and max 16, so to know the adaptivity is working at its best, with areas at min and areas at max sampling rate quite evenly distributed), and you’ll be more able to move towards less time, or more samples, with more precision and speed.
but could you explain more exactely wich noise threshold you mean? AA or DMC? also the “max subdivs” AA or DMC?
I used the matrix exposimeter.
what does “bias” at 0.5 mean? do you have a helpfile for the scripts? I am very interested, but I don’t know where to learn…
I thought decreasing the clr threshold in ADMC means it never reaches 16? I thought it’s like: 1/4@0.01. 1/820.005, 1/16@0.002 etc.
when using the matrix exposimeter, do I have to setup for final rendering, before I use it? do I have to leave the DMC to default?
Well, I render loads of crops with no AA and very low material subdivs XD
Jokes apart, the idea of extending the exposimeter to take into account the sampling rate was sparked from the personal need to see numbers expressing the quality of the sampling, rather than use my susceptible eye to judge noise levels.
The BIAS parameter is basically the average brightness of the sampling image (the bluish one).
If max brightness (1.0) is max sampling, and min brightness (~0.0) is min sampling, an average of 0.5, in an ideal (ie. not of this world) image, means the sampler is sampling equally in the min and max throughout the frame.
In other words, some parts will be at max sampling, others at min, and the average of them will be in between the min and max value.
This of course is arbitrary, as I have no way of knowing what’s in the image, or what’s the quality the user expects from it.
Still, using this tool in a Universal settings environment (ie. default material/light subdivs, most of the work done by the DMC AA with quite high max subdivs) will give clear hints (and for that matter converge to the bias value given some iterations) of what the sampler is doing, both visually (in the preview) and numerically (% of max sampled pixels, average sampling, and so on), to allow the user to adjust precisely the noise threshold (for AA only!).
Raising the BIAS will mean there will be more of the screen sampled at max sampling, once the script has converged, and it’s also true the opposite.
No preparation should be needed in the scene, and the choice of the three “modes” in the top panel will not affect that pass, only the exposimeter pass if it’s enabled (you could run only the sampler pass, if you un-check the expose pass).
Okay. I loaded up ur script with a basic scene. teapot & plane ( with blurry reflections) & Hdri dome light. If I understand correctly the script tells how good ur AA settings are working and changes ur AA thres to average out the subdivs used within the settings that you’ve chosen??
lets just say I this was going to be an animation. and I wanted to work with unclamped 32bit RPF’s. ( do you work in 32bit unclampled or something else? )
Aside from rerendering over and over again until you reach the quality you want ( lets assume TV/cinematic production quality ) Is there good rule of thumb I can start at? or some other trick to finding good min max AA settings? I read a PDF my user Brian541( sp? ) and he recommended never going over 6max 4min subdiv, but I’ve also read this doesn’t apply to animation.
First of all, sorry for taking ages to reply. The forum is still sending me a mail in ten, it seems, and i miss stuff along the way. this post was one.
Let’s see if I can clear this up.
It’s no chance at all that the tool is inside the the Exposimeter: the job it does is IDENTICAL to exposing.
Only, it’s performed on the sampling pass (“show samples” in the AA settings, or the proper render pass, as you wish).
And yes, you’re entirely correct: it does ATTEMPT (in that it needs a few iterations to nail it) to match the average brightness of the sampling picture to what you have set in the BIAS.
Whatever is in the scene, and whatever the average brightness, each average represents exactly a certain number of rays traced.
Take two very different shots, with identical min and max AA subdivs, match their sampling to the same average (or BIAS), they both will have traced the same amount of AA rays, invariably.
lets just say I this was going to be an animation. and I wanted to work with unclamped 32bit RPF’s. ( do you work in 32bit unclampled or something else? )
Always full float, yes.
Aside from rerendering over and over again until you reach the quality you want ( lets assume TV/cinematic production quality ) Is there good rule of thumb I can start at? or some other trick to finding good min max AA settings?
Well, it’s entirely down to the scene and exposure, unfortunately.
Brighter pixels are harder to AA than darker ones; finely detailed textures are more difficult than large, even areas and so on.
I read a PDF my user Brian541( sp? ) and he recommended never going over 6max 4min subdiv, but I’ve also read this doesn’t apply to animation.
Needless to say I’m a little confused.
I bet you are: I disagree entirely with the above sentence.
First because “never” is a word I loathe and avoid using.
Second, because doing so (min 4 - max 6) means the AAser will oversample even areas which needn’t oversample (thanks to the forced min value), while it will not be able to clean areas which will need it badly (set a material with a bright glossy reflection of 0.1, and try and clean that with the default subdivs in the material and 6 max subdivs in the AA).
On top of that, all the hard work Vlado put into making the sampler adaptive is thrown out of the window, along with his “universal settings” method, the single biggest eye-opener for me on how VRay works.
I hope I haven’t confused you further.
If that’s the case, feel free to ask for more confused answers
No you’ve definately cleared a few things up! But I have more questions.
lets say I set min1 and max 10 for AA ,use the script and find thats it using the max value a large percentage of the time. how do i know how much I need to raise it?!
I’m still interested to see morbids renders though!
I’ve just recently had switched over to 32bit rpfs so I’m I’m still trying to figure out how to go about finding good AA settings.
