View Full Version : Anti_aliasing with Self Illumination
Camby1298
09-16-2008, 04:20 PM
Anyone figured how to anti-alias a self-illuminated material whose whiteness is above 1?
BTW I rendered the scene with a min/max sampling value of 1/16
Thanks!
Ethan_Janssens
09-16-2008, 05:34 PM
since when does self illumination have anything to do with anti aliasing?
correct me if i'm wrong: the aa depends on the geometry size + angle torwards the camera
that ligt is pure white, so aa will be more visible cause it's an extreme value
first of, a nice mask of aa is to use glare combined with the self illumination in this case :cool:
other then that, why don't u use the region-rendering capacity on those areas with higher aa settings?
another way to reduce bad-aa is by increasing the render output size, and re-shrink it back to the size you want if necessairy
also, personally, i don't understand why, but the concept is easy, if 1/16 isn't enough, try 4/16, if that isn't enough try 16/64 etc etc...
the smaller the render size, the higher you aa settings have to be
aa = pixel / geometry - color related, not material-shader related
Dave_3d
09-17-2008, 03:34 AM
This is something I have encountered also.
It took some time to find out that it is probably a bug (logged and discussing with Autodesk). The self-illumination when it burns intensely (as you say when it is greater than 1.0, sometimes much more intense) causes a break-down in anti-aliasing when you are using 32 bit Float buffer . It is NOT related to initial Anti-aliasing settings (16/4 should be enough for almost all occasions).
Try using 16 bit Integer buffer.
Please let me know if that helps. It will help when I can report others with the same problem.
The downside is you wont be able to save out your render as an EXR or HDR for further manuipulating.
Dave
nisus
09-17-2008, 09:21 AM
if 1/16 isn't enough, try 4/16, if that isn't enough try 16/64 etc etc...
This really sounds like bogus to me. Yes, the trick would work, but at what expense? It's insane to render at such a high AA setting like 16/64... Rendertime would simply go nuts.
An AA of 1 16 is enough most of the time, very rarely one can use 4 16
If you wanna have 'better' AA, play with the spatial contrast and 'bias' your basic setting more to 16 than 1. It's that simple, without the waaaaaiiiiiiiiiiititttt ,-p
rgds,
nisus
Ethan_Janssens
09-17-2008, 11:15 AM
Yes, the trick would work, but at what expense?
ye k true... it's a "noobish" way to deal with the problem.
MasterZap
09-17-2008, 12:48 PM
No, massive oversampling actually does *not* help.
Here's the thing: In a LDR space, the brightest thing is "white". If some object covers 25% of a pixel, it is anti-aliased to 25% of it's color. 25% of "white" is 25% gray, so it will be anti-aliased as one would expect.
In HDR space, there is no limit. A pixel may be 20 times white. And if 25% of that object covers a pixel, 25% of 20-times-white is still 5-times-white. So the pixel will NOT look partially filled, it will look totally white. And no amount of oversampling will change that.
The only solution is to clamp pixel values *prior* to filtering down to pixels. But this loses any overbright info. So you are caught between truncating the HDR-ness of your render and get nice anti-aliasing... or keeping your render fully HDR, and have less anti-aliasing at the edges. It's mathematically inevitable.
If you know you won't re-expose the render in post, you can just clamp it with mib_lens_clamp shader. But it will completely lose any "above white" information.
/Z
Camby1298
09-17-2008, 12:53 PM
The problem isnt with the samples of AA (whether its 1/16-4/16-16/64 (which is insane) and Ive tried different sample setting and spatial settings), the problem lies within AA between pixels where one pixel is higher than the value of 1.0
http://forums.cgsociety.org/archive/.../t-543646.html
This topic has been discussed and Zap offers a few solutions to it.
Speaking of which, anyone use the Simple Exposure Shader? Is it hidden where can I find it and where do I apply it? i.e. The Camera Lens Global Switch?
-Whoa, I didnt see your post there Zap, then Im sure you would know of what you were talking about in the CGSociety Thread above :oP. Thanks for the solution(s)
Still Having trouble finding the Clamp Shader you speak of; is it hidden? Or would it provided by 3rd party? Or is it just under my nose and Ive overlooked it???
Dave_3d
09-18-2008, 03:35 AM
Zap,
If you apply a glare shader to the lens, is this calculated in 32 bit space ?
If so, we could probably use that to smooth the apparent transition effects when it is anti-aliasing from super white to grey.
Is there a way to apply the glare to single objects ? I sometimes find other areas of a scene flare up when it is applied globally.
Dave
Ethan_Janssens
09-18-2008, 03:54 PM
@ nisus:
play with the spatial contrast and 'bias' your basic setting more to 16 than 1
how?.. this concept is new to me, how can I altern the way mental ray "choses" the pixel-detail? I didn't know this was possible? Could you link some info regarding this phenomenon?
greets
nisus
09-19-2008, 01:37 AM
Hi Ethan,
Could you link some info regarding this phenomenon?
I believe the manual talks about this.
rgds,
nisus
MasterZap
09-19-2008, 04:35 AM
Zap,
If you apply a glare shader to the lens, is this calculated in 32 bit space ?
Yes. That is pretty much the "canonical use of Glare", and is pretty much what would happen in a real camera.
If real cameras were *perfect* and really only took the light arriving at a given CCD site into account for a given pixel, the *same anti aliasing problem* would exist in real world digital photography. But due to the fact that light "leaks" laterally in the CCD (as well as is scattered slithgly by the lens) you get point-spread effects in the "film plane", also known as "Glare". These effects actually cover up the "aliasing problem" nicely.
/Z
vBulletin® v3.6.8, Copyright ©2000-2013, Jelsoft Enterprises Ltd.