[M4IF Technotes] Object questions

[the_ether]

> One thing should be clear: ALL ELEMENTS IN THE SCENE ARE OBJECTS.
> Also the sound, also rectangular video, also any graphics and
> text.

Just to be absolutely sure I have understood. Does the maximum of 32 objects
in a scene include things like sound? This implies that a video encoded
using simple profile, level 0 could not have any audio .
Just out of interest, how is stereo sound treated? Is it one or two
'objects'?
> When you do do an analysis,
> it is completely up to you how you divide the srource material
> into objects. This is not normatively described in the standard.

I realise this but was wondering whether there were any practical pointers.
Depending on the method used to identify distinct objects in a scene, I may
end up with hundreds. I was wondering whether there were any pointers, based
on experience / research that would make a decision process to prioritising
which of those 100 objects should be coded as distinct objects and which
should be aggregated into one.
Re: BIFS. I thought that was best for synthetic images whereby the meshes
are being transmitted. If I had a scene with a woman walking across a field
whom I had isolated as a separate object, would I use BIFS or use motion
compensation? (I'm assuming that I had not been able to construct any 3D
model of the woman and all I have is the 2D representation of her).
Re: updating of object definition. So all I need to do is simply define a
new shape bitmap for the new object, I do not need to signal the decoder
that "object 1" is now a different object?
I can only imagine doing this with an I-VOP as a P-VOP would be relative to
the previous frame. If the previous object was a duck and the new one a cow,
then there's not a lot of similarity. A goose on the otherhand...
----- Original Message -----
From: "Rob Koenen" <rkoenen   intertrust.com>
To: "'the_ether'" <the_ether   btinternet.com>; <Technotes   lists.m4if.org>
Sent: 13 November 2023 18:28
Subject: RE: [M4IF Technotes] Object questions
> > 1- What's the maximum number of objects that can exist in a
> > given frame and
> > does this number vary with the profile?
>
> Max objects is per scene, and varies with Profile _and_Level_.
> In Simple   Level 0 it is 1 single object; goes up to 32 for
> the highest levels.
>
>
> > 2- Are there any pointers on how to decide which parts of an
> > image should be
> > treated as objects and which as part of the background? I
> > suppose factors
> > could be size of the object, speed of movement and level of
> > contrast between
> > it and its surroundings.
>
> One thing should be clear: ALL ELEMENTS IN THE SCENE ARE OBJECTS.
> Also the sound, also rectangular video, also any graphics and
> text. Your question seems to relate to encoding however.
> One more thing to be clear YOU ARE NOT FORCED TO DO AN ANALYSIS
> when you want to encode a sequence. Just can just treat the whole
> scene as a singular rectanguler object. When you do do an analysis,
> it is completely up to you how you divide the srource material
> into objects. This is not normatively described in the standard.
>
> >
> > 3- Can I arbitrarily change which objects I am using (ie
> > whereas object 1
> > was a tree in frame 1, in frame 2 I decide that the tree no
> > longer needs to
> > be treated as an object and so I allocate object code 1 to a
> > cow instead)?
> > Or can I only make a change if the new object allocation occurs in an
> > I-frame?
>
> You can arbitrarily change objects and update the scene using BIFS.
> If you decide to merge the duck into the background OBJECT, you
> need to update that object. If you want to do that using an I-VOP,
> fine. If you think you can do it with a P-VOP, that's your choice.
> Individual objects can have their own 'object rate'.
>
> > 4- Can objects have holes in them, like a doughnut or are
> > they assumed to be solid?
>
> Objects can have lots of holes. You can actually use an alpha mask
> to describe the object, so it can even have fuzzy edges.
>
> > 5- I noted that I can transmit object information with or
> > without motion
> > vectors and also with or without prediction errors. What if
> > the shape and
> > position hasn't changed. Do I transmit anything at all and
> > assume that the
> > decoder will re-draw what it had last time?
>
> This is no different than in 'normal' video coding.
>
> Rob
> _______________________________________________
> Technotes mailing list
> Technotes   lists.m4if.org
> http://lists.m4if.org/mailman/listinfo/technotes
>