PDA

View Full Version : Multichannel OpenEXR


suparx
06-14-2008, 07:27 AM
does anybody know if xsi can render a scene with all its passes/channels into a single OpenEXR file?
i've tried several plugins but either they didnt work for me or they made me dependent on their own shaders.

it would be great if softimage could support OpenEXR file format much better in future.
i guess this format will clean up workspaces sooner or later.

ThE_JacO
06-14-2008, 09:10 AM
As far as I know the lack of support for additional channels is an issue with MRay not with soft.

As for multichannel, personally I'm not that interested, and the support compositing side is still so minimal that I wouldn't be able to use it on 80% of the jobs anyway.

Even if it was there I'd still render to separate images and then eventually merge them into multichannel later.
Single files with multiple data sets are a sure fire way of ****ing yourself up, not to mention they are impossible to frame check through automation, which makes them completely useless from a farming point of view (and with impossible I of course mean highly bloody impractical :) ).


Other than multichannel openEXR support in XSI in the last year has finally reached maturity and reliability, I haven't had any problems with it personally.

suparx
06-16-2008, 05:52 PM
As far as I know the lack of support for additional channels is an issue with MRay not with soft.

As for multichannel, personally I'm not that interested, and the support compositing side is still so minimal that I wouldn't be able to use it on 80% of the jobs anyway.

Even if it was there I'd still render to separate images and then eventually merge them into multichannel later.
Single files with multiple data sets are a sure fire way of ****ing yourself up, not to mention they are impossible to frame check through automation, which makes them completely useless from a farming point of view (and with impossible I of course mean highly bloody impractical :) ).


Other than multichannel openEXR support in XSI in the last year has finally reached maturity and reliability, I haven't had any problems with it personally.

hey raffaele, thanks for your reply.
do you like using passes more than channels then?

sure the compositing side has still a lack of support for openexr.
for compositing i personally tried ps, fusion and ae. they all work fine
if you have the right addons installed.

i think for small to middle size productions it would be a great option.

skyphyr
06-17-2008, 04:39 AM
It's a mistake to think passes and channels are interchangable. Channels should be used for information that is already being calculated or cannot easily be calculated as a separate pass. So good examples for using channels would be depth, motion vector, reflection, albedo, diffuse and specular illumination. A good example for pass is ambient occlusion and separation of elements (i.e. rendering the foreground/characters separate from the background so that once your camera is locked off if character animation changes you've greatly reduced the amount of re-rendering requried).

I'm personally not particularly fussed on using multiple channels for openexr. It's not the primary benefit of openexr by a long shot. It's far more useful to be working and storing your render in linear space and float is the best option for this to ensure you're not losing data.

Cheers,

Alan.

CiaranM
06-17-2008, 09:37 AM
I would also add that I've found channels to be particularly useful with shots containing heavy displacement, SSS, or other elements that require long pre-processing. In such a situation, you can get all your data out in a single pass if you do it right. But, you'll need to set up a system of overrides in case you need to go in and re-render a single one of those channels...
I've also heard that lack of multi-channel EXR is a Mental Ray problem, but I don't think I'd take a gamble on using it, even if it was supported.

ThE_JacO
06-20-2008, 01:14 PM
Channels in MRay and channels in openEXR are completely different things.
Channels as in MRay channels (AOV aka Arbitrary Output Variables in renderman engines) are useful because you can dump buffers as raster images on disk from a single process chain (shader/phenomenon). Passes serve a different purpose and require a re/load and reprocess.

Channels can be comfortably output as of now into separate images, and XSI's way of dealing with it (tokenized output structure) is both smart and easy to use.

Channels in openEXR are just an aggregation utility that allows you to have several "layers", or different raster images, contained in the same storage identity (the file on disk).

Having support for multi channel in openEXR or not has virtually no effect whatsoever on rendering out channels or passes, so the two things are completely separated.
Leaving MRay channels/passes out of this discussion, it's multichannel openEXR that I would find pointless, and wouldn't use, even if it was available.
The hyperbolic version of why I don't like multichannel is the same reason why I render frames instead of already assembled movies.

In design pattern theory artificially merging data at output time, when it's computed and saved separately anyway, brings no benefits whatsoever other than ease of management and packaging.
XSI's channel management and a decent directory structure already cover completely efficient retrieval, so I would NEVER add packing to that process and go for multichannel openEXR because I can see no benefit in it, but lots of issues and disadvantages (ESPECIALY for the small scale production).

If you really can't deal with not having it you're still better off running packing operations in batch with some compositing software that supports it, with the thought in mind that the very moment you do that you render your data absolutely useless to the vast majority of software on the market.

In short: openEXR is already supported to the extent it needs to be. Output is good, files are well written, all depths are available and there are no odd metadata and header issues.
And to answer your question: I would use openEXR, and have in fact used it exclusively for the last three years. It doesn't need to be "better supported" as it already is. Multichannel is, in my opinion, a waste of time and not needed at all for me to decide using EXR.

Maybe adding a bottom to top option to the saver would be nice to make shake happy, as it would be having lzw and pyramidal TIF support, but that's about it as far as file I/O goes.

Kel Solaar
08-11-2008, 03:05 PM
pretty much the same as ThE_JacO here, there is a lack of flexibility with having all your passes in the same file, imagine you need to recalculate only one framebuffer because there is something wrong with it :|