Hi Carsten,
On Thu, 30 Oct 2014 18:38:42 +0100
Carsten Kurz <audiovisual(a)t-online.de> wrote:
Hi Carl,
I have received a video in AVI container, DV codec, so, pretty
standard. However, it is anamorphic, so no square pixels. Various
media players on my machine detect it properly as 16:9, but
DCP-o-matic 1.76.7 detects it as 720x576 (1.25:1) and squeezes it
when choosing 'no scale'. No big deal to correct this manually in the
scaling options, but maybe you can fix that. Since the media players
display it correctly, there must be a flag the DCP-o-matic currently
ignores when analyzing it.
The same happens when importing 16:9 VOBs from e.g. DVDs.
As you have observed, DCP-o-matic really thinks of everything as having
square pixels. I like the simplicity of this, especially as I'm not
entirely convinced that we can rely on pixel aspect ratio values in
video files. I have no evidence to back up my doubts, though.
Hmm, I understand that there is some interpretation
necessary.
DCP-o-matic first follows Default Scale To in prefs. While that is
correct, and some options chosen have to result in wrong aspect ratio
scaling, I think that at least the 'no scale/no stretch' options
should follow the original content AR, if it is clearly indicated in
the file metadata. At least the content description should indicate
that the files is flagged 16:9/anamorphic.
Aside from what the fact that in this case I can create a proper
upscaled video - there is no way I could create an unscaled version
of this file with the correct aspect ratio. When I choose 16:9, it
get's uprezz'd, when I choose unscaled, I get a square 720/576 in a
flat frame, when I choose NoStretch, I get a square 1350x1080. There
is no way I can assign a 16:9 reference to this file.
How could you put an unscaled version of this file into a DCP at all?
If my maths is right the pixel aspect ratio is 1.42:1, so aren't you
going to need "upscaling" of some description to map it to the
square-pixel DCP?
I think the proper way to handle this is to find a
16:9/non-square
pixel flag in the file and act accordingly, that is, display it
stretched and with a 16:9/anamorphic indication.
I think that fight for all necessary yet straightfordward arranging
of scaling options will never end...
I'll have a think about it. My concern is that I think I would rather
come down on the side of simplicity of the processing chain (i.e.
thinking of pixels as square) over getting the required scaling
settings absolutely right first time with no input from the user.
Maybe we should ditch the default scale-to option and instead guess it
from the input AR and pixel AR? I guess that would be extensible if
and when DCP-o-matic tries to detect letterboxing/pillarboxing of
inputs.
Best regards,
Carl