ART/SDK Feature Request: Better Handling of Anamorphic Footage

[KrisZ]

Hello,

I tried submitting some feedback to the ART survey on the webpage, but figured I'd post this here as well with more detail.

I would like to see a better, more consistent, clearly-labelled output mode for decoding full-resolution anamorphic footage, with correct pixel aspect metadata, in ART and the ARRI SDK. In my case it's for VFX purposes.

For example, imagine you have some footage shot 2x anamorphic, open gate, ARRIRAW on ALEXA 35. The current behavior of ART 1.5.0 on Mac is as follows:

By default it will recognize the anamorphic nature of the footage and set the lens squeeze to 2.00, and if you then use the output resolution mode "As Input", it will want to scale the image down from 4608x3164 to 4608x1582 de-squeezed with square pixels. To begin with, I'd like to say throwing away half the vertical resolution is not what I'd consider "as the input," so I find that option a bit of a misnomer, and it's not something you'd generally want to do at this early stage in the VFX workflow. But regardless, if you export it using ProRes or TIFF, then the footage will be what you'd expect. However if you export OpenEXR, then it still sets the pixel aspect metadata of the image (exr/pixelAspectRatio) according to the lens squeeze. So if you bring in those exported EXRs to Resolve or Nuke, the footage will look double de-squeezed, once from the downscaling/de-squeeze in ART, and once from the pixel aspect meta.

Conversely, if you want to retain the original, full-resolution image with non-square pixels, you have to manually override the lens squeeze to 1.00, and then again the output will differ depending on the format. ProRes/TIFF will skip adding any pixel aspect meta and you have to manually set it correctly in the application you import the footage to, if you want proper anamorphic behavior. Or if you use OpenEXR, then everything will behave correctly, thanks to the pixel aspect meta still being set according to the lens squeeze.

The last result with EXRs is the output I'm looking for, except I'd like it to be explicit and consistent within ART/SDK, maybe with a simple option/toggle from the current behavior. Changing the lens squeeze doesn't seem right and is needlessly confusing. I still want to use the lens squeeze factor, just as a pixel aspect instead of for actually scaling the image during decode. So in this proposed alternative mode I'd like to see:

• For the "As Input" output option to really do what it says, and not change the resolution or scale down the image.

• For the detected lens squeeze to determine what pixel aspect metadata to set, not how to scale down the image.

• Consistent behavior between different output formats. Not sure if TIFF supports pixel aspect metadata so maybe that one can be ignored, but the others do.

• Some UI feedback to make it clear what's going on with the decoded output. So instead of just "4608x3164 -> 4608x3164" in the overlay, maybe "4608x3164 -> 4608x3164 (pixel aspect 2.0)" or something like that, maybe also keep the de-squeezed video preview.

Now the reason for this isn't that I want to process all footage using ART. However I have previously been in contact with The Foundry who are implementing the ARRI SDK in Nuke, and they are understandably using the ARRI Reference Tool as a reference, and are either reluctant to have behavior that isn't reflected in ART, or unable to change the behavior of the SDK. So I guess the issue is that the reference is a bit unclear and inconsistent. In practice this means when bringing in such anamorphic footage to Nuke directly, you have the same non-ideal options: either let the detected lens squeeze cause the image to be downscaled/de-squeezed (which as mentioned, you generally don't want for VFX at this point), or manually override it to 1.00, which skips setting any pixel aspect metadata (like on the ProRes exports of ART), meaning you have to manually add it back to get proper anamorphic behavior. This is quite bothersome and necessitates extra pipeline code and workarounds.

If it's not possible to add this alternative mode, would it at least be possible to get clarification on whether, when overriding the lens squeeze to 1.00 in the decode options, the pixel aspect metadata of the image should or should not still be set according to the detected lens squeeze? Since the current behavior isn't consistent between OpenEXR and ProRes. If it should—and I definitely think so, because ignoring it doesn't make any sense in most cases—then maybe I could at least take that information back to The Foundry and have them properly set the pixel aspect metadata when overriding the lens squeeze decode option.

Thank you.