[Matroska-devel] Proposal for a better definition for the 'keyframe' flag

Jan Ekstrom jeebjp at gmail.com
Sat Jun 15 16:01:58 CEST 2013

On Sat, Jun 15, 2013 at 3:12 PM, Steve Lhomme <slhomme at matroska.org> wrote:
> When do you see the second definition in the specs ? I cannot find it.

http://matroska.org/files/matroska.pdf (as I noted, the Matroska
'specification' PDF), page 39. Since the definition on the web page
isn't exactly definitive, it leads to people looking for other
definitions, like this one.

> This SimpleBlock contains only random access point frames. When decoding
> starts from a random access point, it can be correctly decoded, as well as
> all the following multiplexed data in presentation order, without having
> access to previous multiplexed data in coding order.

I'm not sure if this is noted anywhere else, but as far as I can see
this limits the SimpleBlock to contain a single "sample" (MPEG-4 Part
12 lingo), if this is how SimpleBlocks should be used, then this
should be defined somewhere else as well, methinks?

Also I am not sure about the wording of "contains only random access
point frames", "This SimpleBlock consists of frames that form a random
access point" is what I just came up with some thinking. A random
access point is just a tag, you should not, if possible, limit its
contents too much. Unless you want to define a whole lot of extra
things like the formal part of 14496-12's random access point
definition does, that is.

For reference, the 14496-12's random access point definition:

random access point (RAP):
Sample in a track that starts at the ISAU of a SAP of type 1 or 2 or 3
as defined in Annex I

NOTE: Informally, a sample, from which when decoding starts, the
sample itself and all samples following in composition order can be
correctly decoded.

Jan Ekström

More information about the Matroska-devel mailing list