openexr-user
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Openexr-user] compositing with EXR


From: John Coldrick
Subject: Re: [Openexr-user] compositing with EXR
Date: Fri, 28 Oct 2005 09:36:14 -0400
User-agent: KMail/1.8.2

On Friday 28 October 2005 05:26, ant l wrote:
> Thanks for the reply..I'll try to explain better.
> I understand that exr is just another file format and that it can be
> treated as such. However what seems unclear to me is how this is used in a
> production pipeline.
> Typically, cineon are converted to linear format and stored in 16bit
> images.
> OpenEXR images are read as 32bit floats (by Shake at least) and are 
> quite slow to work with.

        Actually I've had good speed response with OExr in Shake, I suppose 
it's all 
relative.  It's trivial to convert to 16, of course, if you want, but just as 
you would typically want a depth map to be float, OExr gives you that much 
more to fiddle with.  It's options.

        Houdini's compositor, for instance, keeps OExr images at 16 bit float 
internally.  AFAIK Shake comps at float internally.  It's where it all ends 
up eventually, which is why I'm a little curious why you're finding such a 
noticeable speed hit. I find it *far* snappier on an NVidia FX3600 than 32 
floats(probably due to using the clever nvidia half display hack which was 
one of the great things about OExr - see the website).  Graphics driver 
issue, perhaps?  I tend to render CG layers at 16 float in OEXR just for that 
reason, since internally the renderer is already doing it at pure float and 
dithering it down to whatever depth you want.  OExr offers that great option 
of depth yet excellent filesize performance.

> Is this a price that 
> we have to pay? I tried to convert them to 16bit (to match the cineon
> plate) however, they get clipped to 1 which makes it a useless operation.

        I'm not sure I'd call it useless, really.  Getting values outside black 
and 
white are very useful for some film shots, but even without that you're 
getting higher depth within 0-1.


> I guess my question is if people generally use them as 32bit floats or
> convert them to something smaller? And if so, how?

        Personally, I'd only convert in the pipeline, e,g, using the Bytes node 
in 
Shake.  Sometimes I find what I want to do in a comp needs it from 0 to 1, 
otherwise I need to manage it, so for a certain operation I'll nip it down.  
Otherwise, I like the "room".  :)

        Cheers,

        J.C.

-- 
John Coldrick                  www.axyzfx.com        Axyz Animation
Houdini/Renderman/Discreet                           425 Adelaide St W
416-504-0425                                         Toronto, ON Canada
address@hidden                                        M5V 1S4
-----------------------------------------------------------------------
It's a damn poor mind that can only think of one way to spell a word.
                -- Andrew Jackson




reply via email to

[Prev in Thread] Current Thread [Next in Thread]