[Mp4-tech] [video] [H264] IDR-pictures, stream splitting and conformance

[J. Miles]

Hi experts,
I have a number of questions on the use of IDR picture in relation to
stream-splitting and random access and conformance to the H.264 / MPEG-4 AVC
Standard that I hope someone can help answering:
Assuming we have a H.264 coded bitstream that is in conformance to the
H.264Standard and that we have somehow stored this stream (not really
relevant).
The bitstream thus consists of one or more coded video sequences each
beginning with an IDR picture and followed by any number of non-IDR
pictures. If such a stream needs to be divided into smaller parts that are
independently decodeable of each other, would not this have to be done at
the coded video sequence level to ensure continued conformance of each
partition? If we later want to decode the partitioned stream without the
full history of the entire bitstream, we need that each partition is still
conformance to the Standard. In other words, splitting into smaller
independent parts can only occur between video sequences (before IDR
pictures)?
Having a bitstream that consists of a single very long video sequence can
not be divided then without breaking the standards-compliance of the
individual partitions. That is my understanding but is it correct?
Apparently, the divided partitions when accessed individually would have to
include any non-VCL NAL units (e.g. sequence and picture parameter sets)
delivered by some means to ensure that the video sequence can be properly
decoded.
The decoding process, I assume, is normally started at the beginning of the
bitstream. But then the Standard also mentions 'random access' (mostly
within section D.2.7 on SEI recovery points) that is defined in 3.112 as
"The act of starting the decoding process for a bitstream at a point other
than the beginning of the stream". This is actually what we want to do with
the divided stream parts: random access but under full compliance with the
Standard. When the Standard mentions starting the decoding process at a
point not at the beginning of the stream and the related problems with that
(e.g. missing references), it does not refer to that the stream could just
be divided at any reasonable place and still be compliant, or..?
I hope that the above is sufficiently clear to understand. But what I am
basically asking is: What is required of and how can a compliant H.264 video
stream be divided into smaller parts that are still compliant, and how
should a compliant decoder handle such divided streams..?
Thanks.
Jay
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20080310/24683d3a/attachment-0001.html