Yes, the answer is technical because that's simply the nature of the problem. As much as we'd like to believe we can be shielded from the gory technical details of working in 3D, they do have a way of making their way up to the surface.
These errors are extremely tricky to track down, which is why I asked people to send repro scenes at the bottom of the email. The many repro scenes I received because of that email, I used to do a major pass over our shader libraries for 5.11 to ensure that this effect would, if not eliminated, at least be minimized. If you still have scenes with black spots in XSI 5.11, please send them to either me directly or to support(at)softimage.com.
Unfortunately, there are some closed libraries from mental images (notably the sub-surface scattering) and/or 3rd parties that we have no control over. I've reported what I can to mental images and they've responded with some fixes. Others I've tried to pass on to 3rd parties as much as I've been able to dig up an email address for them.
In previous versions of mental ray they would convert the NaNs to red/white samples (depending on version) which would then get filtered into the final pixel. They would then show up as weird flickering in the image. Now they simply let the NaNs go all the way to the filtering stage which is why we now get black spots instead (the bigger your filter size, the larger the black spots).
I'm sure some kind soul could whip up a camera shader that simply throws away those NaN samples. It would overall slow your renders but at least you wouldn't end up with black spots. Alan?
- ½
-----Original Message-----
From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM] On Behalf Of Kris Rivel
Sent: Wednesday, August 09, 2006 10:30 AM
To: XSI(at)Softimage.COM
Subject: Re: flickering with zbump
Well that answers nothing for me or the typical artist. I'm not going
to build my render tree thinking about NaN values or floating point
inaccuracies. I understand it somewhat but the everyday artist probably
won't. My biggest problem was that these never appeared in previous
releases so I'm thinking its either the latest version of MR or a recent
version of XSI that is the culprit. It just happens way too often in my
opinion.
Kris
Bernard Lebel wrote:
> Halfdan actually answered that question on the mr list:
>
>
> "It's a combination of mental ray, the shader libraries and floating
> point inaccuracy.
>
> A shader returns NaN (not a number) value which overwhelms any other
> value in the array of samples which are filtered down to a pixel. If
> you increase your filter size, for example, the black spots get
> bigger.
>
> When a NaN is converted to 8- or 16-bit value, the conversion function
> returns zero, which is why you see black spots. If you save the image
> out as float, the NaN values will be saved out with it.
>
> This is usually caused by a shader passing a negative value to sqrt().
> A simple example:
>
> nd = mi_vector_dot( &state->dir, &state->normal );
> val = sqrt( 1.0f - nd );
>
> Under some circumstances, even if both state->normal and state->dir
> are a supposedly unit vectors, the dot product can give a value
> slightly larger than 1.0, due to floating point inaccuracies. This
> then gets inverted and passed as a very small negative number to
> sqrt() which then returns a NaN. This NaN value then proceeds to
> contaminate any and all calculations it is involved with downstream.
>
> http://en.wikipedia.org/wiki/NaN
>
> These kinds of errors are very often highly non-trivial to uncover. My
> recommendation is to talk to your vendor. The more examples they get,
> the easier it is to track these things down."
>
>
>
> Cheers
> Bernard
>
>
>
> On 8/9/06, Kris Rivel <kris(at)krisrivel.com> wrote:
>>
>> What's with those damn black artifacts? Can anyone at Soft give a
>> clear
>> explanation of the problem and maybe put as at ease and tell us it
>> will be
>> fixed in the next release. It happens way too often and there
>> doesn't seem
>> to be a clear repro for it. I know I've submitted some nasty scene
>> files to
>> illustrate it. The problem is that its somewhat random but seems to
>> happen
>> often when using OpenEXR, HDRI or shaders like zbump or dark tree. I
>> had it
>> pop up again on a job where I was using an OpenEXR image as an
>> environment
>> for reflections. I got lots of the little buggers so I converted it
>> to HDRI
>> and it worked fine. It just happens a bit too often I think. If its
>> an MR
>> thing, we need to bash their heads in a bit to address the problem.
>>
>> Kris
>>
>> kim aldis wrote:
>> Didn't the new version fix that?
>>
>>
>>
>> -----Original Message-----
>> From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM] On
>> Behalf Of adrian
>> Sent: 09 August 2006 12:20
>> To: XSI(at)Softimage.COM
>> Subject: Re: flickering with zbump
>>
>> we had issues with this, but it manifested itself as 2x2 or 4x4 pixel
>> black squares.....
>>
>> resolved by lowering the bump factor, or some convoluted rendertree
>> chicanery to stop zbump returning NAN values
>>
>> not much help i know
>>
>> a
>>
>> Adrian Wyer
>> Fluid Pictures
>> 33 Glasshouse Street
>> London
>> W1B 5DG
>> ----- Original Message -----
>> From: "Chris Marshall" <chris(at)eclipsecreative.co.uk>
>> To: <XSI(at)Softimage.COM>
>> Sent: Wednesday, August 09, 2006 12:22 PM
>> Subject: flickering with zbump
>>
>>
>>
>>
>> Hi All,
>> Does anyone get flickering problems when using zbump?
>> I've seen it on two jobs now, and it's so bad we've had to go back to
>> regular bumpmapping, which isn't ideal.
>> It happens when you have an object breaking through the surface of
>>
>> the one
>>
>>
>> using zbump. The flickering occurs around the area where the one
>>
>> object
>>
>>
>> penetrates the other.
>> Anyone seen this or can offer some advice on fixing it?
>>
>> Thanks
>>
>> Chris
>>
>>
>> ---
>> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
>>
>> body:
>>
>>
>> unsubscribe xsi
>>
>> ---
>> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
>> body:
>> unsubscribe xsi
>>
>>
>>
>>
>> ---
>> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
>> body:
>> unsubscribe xsi
>>
>>
>>
>>
> ---
> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
> body:
> unsubscribe xsi
>
>
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi
