RE: FloatingPoint Render Weirdness
| Date : Fri, 6 Jul 2007 11:51:24 -0400 |
| To : <XSI(at)Softimage.COM> |
| From : "Luc-Eric Rousseau" <lucer(at)Softimage.COM> |
| Subject : RE: FloatingPoint Render Weirdness |
In my opinion the result is correct, although not as plug-and-play
as you'd like. It does need to be tone-mapped, and I would
process the image to add a halo to the out-of-range spots (i.e. blur
the highlights).
The AA artifact reminds me of when you point a digital camera to a bright light
source, there's ugly color clipping. So in the same way, I'd look into adjusting
the exposure and also mimic how a lens (or the eye) would react to the
having this really bright spot there. The simple way that AA works doesn't mimic
these physical phenomena, it is only there to control super-sampling and
blend its results equally.
btw, if I'm not mistaken - and "Halfdad" (or "Diaper Dan") will correct me - when
mental ray renders to 8-bit, I think it filters the 8-bit samples and not
floating point samples, so that would explain why you'd get different results.
From: owner-xsi(at)Softimage.COM on behalf of Votch
Posted At: Fri 7/6/2007 3:53 AM
To: xsi(at)Softimage.COM
Posted To: xsi
Subject: FloatingPoint Render Weirdness
I've ran into a little problem when rendering to floating point images in 6.01. I've talked to support and they are looking into the issue but I thought I'd run it by you all as well.
It appears that XSI/MR doesn't evaluate antialising properly in floating point pixels when a floating point file format is selected for the output file type. However when rendering to an 8 or 16 bit format everything renders correctly. It's strange because there is a visual difference when rendering to float formats compared to files rendered clamped.
I've attached a scene that demonstrates the problem. Just render out all 10 frames of both passes. Then compare the frames in a compositor. Frame 1 looks correct because most values are below 1. As the spot light intensity increases the aliasing gets worse in high contrast areas as soon as pixel values go above 1. On frame 10 you can really see the issue.
Support recommended using the ToneMapping lens shader. It works great for squishing floating point data into a clamped file, but it's not ideal because rendered images are no longer linear.
In MR the error shows up with any float file format. All Clamped formats look correct as well.
I also ran tests using this same scene in 3Delight via Affogato without issue. I haven't tried MR in Maya or Max Yet.
I first noticed this problem in creature renders using HDRI environment domes. The reflection and spec always looked aliased in EXR renders but normal in TGA renders. After a lot of investigating I've come to the above conclusion.
Anybody encountered this before? Any Idea's...
Votch Levi
It appears that XSI/MR doesn't evaluate antialising properly in floating point pixels when a floating point file format is selected for the output file type. However when rendering to an 8 or 16 bit format everything renders correctly. It's strange because there is a visual difference when rendering to float formats compared to files rendered clamped.
I've attached a scene that demonstrates the problem. Just render out all 10 frames of both passes. Then compare the frames in a compositor. Frame 1 looks correct because most values are below 1. As the spot light intensity increases the aliasing gets worse in high contrast areas as soon as pixel values go above 1. On frame 10 you can really see the issue.
Support recommended using the ToneMapping lens shader. It works great for squishing floating point data into a clamped file, but it's not ideal because rendered images are no longer linear.
In MR the error shows up with any float file format. All Clamped formats look correct as well.
I also ran tests using this same scene in 3Delight via Affogato without issue. I haven't tried MR in Maya or Max Yet.
I first noticed this problem in creature renders using HDRI environment domes. The reflection and spec always looked aliased in EXR renders but normal in TGA renders. After a lot of investigating I've come to the above conclusion.
Anybody encountered this before? Any Idea's...
Votch Levi
- Follow-Ups:
- Re: FloatingPoint Render Weirdness
- From: Votch <megavotch(at)gmail.com>
- Re: FloatingPoint Render Weirdness
- References:
- FloatingPoint Render Weirdness
- From: Votch <megavotch(at)gmail.com>
- FloatingPoint Render Weirdness
| DATE: | << | >> | THREAD: | << | >> | INDEX: | Main | Thread |
|---|
- Previous by Date: Re: What format and bit depth do you use?
- Next by Date: Re: What format and bit depth do you use?
- Previous by Thread: Re: What format and bit depth do you use?
- Next by Thread: Re: What format and bit depth do you use?
- Index(es):
| Search the XSI List archives here or use the advanced search form to search across mailing lists. Searching help is available. |