See also: IRC log
<trackbot> Date: 17 April 2009
Silvia, you miss some amazing food yesterday night, but you will see the pretty pictures taken :-)
you miss as well the longggggggggggggg drink session :-)
trackbot, start telecon
<trackbot> Meeting: Media Fragments Working Group Teleconference
<trackbot> Date: 17 April 2009
<silvia> any changes to the agenda?
<scribe> Meeting: Media Fragments WG, 3rd F2F Meeting - Barcelona (Spain)
yes Silvia, we will start with the session: Implementation Reports
scribe: then talk about the
... and finally about the test cases, where we need to kick off the work
Does that suit you ?
Silvia, we are waiting for Jack, coming in a few minutes
scribe: and Yves, that he could phone us
<scribe> Scribe: Raphael
<scribe> Scribenick: raphael
Raphael: we should have a picture
representing the general structure of the media object, the H,
H', K, D, etc. we talked about yesterday
... Can we do such a representation valid for most of the formats?
Conrad: yes, for the streamable
... I'm happy to take an action to draw such a schema
<scribe> ACTION: Conrad to draw a representation of the general structure of a media resource, for streamable formats [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action01]
<trackbot> Created ACTION-69 - Draw a representation of the general structure of a media resource, for streamable formats [on Conrad Parker - due 2009-04-24].
a) Davy wants to present and demo a software that allows to slide media files independently of the codec format
b) Jack has some reporting to do on its implemntation
c) Davy wants to talk about RTSP implementation
d) Conrad wants to report on his experience in Annodex
Silvia, Davy will start with some slides that are coming
Davy going through a presentation
Davy: Slide 6, my PhD work is
built on BSD content adaptation, using RDF
... based on a model for media bitstreams encoded in OWL
... Slide 7: Coding formats -> model for media bitstreams -> container formats
... Slide 8: model for media bitsreams represented as a set of classes and properties
... left part is the structure part, while the right part is more semantic
... Slide 9: shows the workflow
... Slide 10: shows an example, two tracks (audio, video) represented in the intermediary model
... assume there is a mapping between one particular codec format and this model
... Track selection, means selection of MediaBitstreams, straightforward
... Temporal selection, means selection of data blocks, based on their timestamps
... problem for Spatial clipping?
Jack: it works only for scalable formats for Spatial Fragment
Davy: Slide 11: general
architecture of the NinSuna platform
... the retrieval service is not yet compatible with the media fragment syntax, but we could adapt it
... Slide 12: two remarks, 1/ is the media needs to be 'ingested', i.e. the intermediary model needs to be instantiated and 2/ currently client decides the output format
Time for demo
<mhausenblas> DON'T CLICK on the following link, now:
<silvia> what is the link for?
<mhausenblas> silvia: that's davy's demo :)
<mhausenblas> note that the demo requires IE and Silverlight
<silvia> ah, thanks
scribe: and VLC for the playback
<Yves> raphael, no
<Yves> we will put thata on another server
<erik> our demo can be found at http://multimedialab.elis.ugent.be/NinSuna/
<silvia> is there a screencast?
A screencast will be made by the IBBT guys next week
<silvia> ok - those screenshots at the demo site are good, too
I'm uploading the videos on my web page
cofffee break, we are back in 15 min
Videos will be at http://homepages.cwi.nl/~troncy/tmp/spool/video/
Silvia, we will resume
Raphael: now we will listen to Jack
Jack: I took the Python URL
... I took this library and add some code of myself
... for parsing the media fragments part
... and here I discover all the issues discovered and reported, now fixed
... and then the problems started, I tried first with the file URL (local resource)
... problem is with synch on video/audio, should we do transcoding or not, etc.
... my goal was to first address: file://myvideo.mp4#t=10,20
... use ffmpeg for slicing the media
... it's hard to use it, the API changes frequently, but the functionalities are very good
... at the C level, it does what we want, but at the command level, it is meant to do transcoding
Guillaume: why not using the Python library G-streams, an API for ffmpeg?
URL lib for Python: http://docs.python.org/library/urllib.html
<guillaume> Gstreamer Python binding library : http://gstreamer.freedesktop.org/modules/gst-python.html
<guillaume> and GStreamer itself http://www.gstreamer.net/
Jack: I will have a working
version in a few more days, but it will transcode
... since this is what ffmepg does
... my plan is thus rather going through C++ implementation, to attack ffmeph low layers, and have no transcoding
... I can share the code that does the parsing of the fragment
<mhausenblas> if not, I might have some resources (not myself, for sure, but a PhD of mine) who could do something based on jQuery
<scribe> ACTION: Jack to commit in CVS (code directory) his python code doing the parsing on client side of the media fragment [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action02]
<trackbot> Created ACTION-70 - Commit in CVS (code directory) his python code doing the parsing on client side of the media fragment [on Jack Jansen - due 2009-04-24].
Conrad will now report on his implementation experience
Davy, first report on RTSP implementation
Davy: RTSP implementation wiki
... quick test, I have an implementation of the media fragment over the RTSP protocol
... it works for track and temporal selection
... first communication is the Describe, and the server answers which track is available, etc.
... then for each track, we have a Setup
... so we can have easily a track selection in RTSP
... next command is the play command, where we can specify the temporal clipping
... I don't remember which units is supported, but for sure npt and smpte
... again, I have an implementation that parses the media fragment, and generates a PLAY command
... i do not yet cover the whole grammar of the media fragment, so far just, t= and track=
... implemented in C
... I'm working on covering the whole grammar
Raphael: my (silly?) idea, embed
VLC in a web page
... implement on client side, in the browser, the possibility to write down a media fragment URI (rtsp), use the code of Davy to generate the good Describe, Setup and Play commands
... send that through VLC and get the results back, using the VLC plugin in the browser
Davy: should be doable
... my issue is with track selection, how do you know the name of the track
Jack: this is related to the
issue raised by Silvia, http://www.w3.org/2008/WebVideo/Fragments/tracker/issues/4
... what I come accoss, is that often, you want not 1 track, but 2 tracks, or all tracks except one
... not very convenient with our current scheme?
Silvia: suggestion to add a comma separation, such as: #track='track1,track2,track3'
Jack: or event having a minus
<silvia> the naming of the tracks is up to the container format - some allow it, others don't
<mhausenblas> Scribenick: mhausenblas
<silvia> for example QuickTime has chapter tracks that should be addressable in the given way with #track
Conrad has essentially said everything yesterday, now no new stuff
<silvia> Jack: mentions that naming of tracks is decided by author and they create the URL
<silvia> Conrad: mentions that with Accept-Language it should be possible to leave the track selection for languages to the server
raphael: conrad ok for you to implement it based on # rather than ?
conrad: will try to implement the hash part as well
raphael: agree, after the first
step both # and ? have the same procedure
... conrad would you implement the 'missing link' using the HTTP Link: header draft
conrad: yes, I like Yves idea using it, gonna try
raphael: Yves, silvia any question?
<scribe> Scribenick: erik
yves: give a set of URI's and describe what should be expected
michael: rather have a set of simple cases to start with
<Yves> parsing URI is more or less done (via the code I contributed), so it's more how a cient parse que URI and act based on that
michael: really stress on what
the outcome is
... try to make it as automatic as possible
Jack: some temporal testing
should at least be able to be done automatically (cfr. via
... some manual video-inspection still has to be done
<conrad> we could add time range validation to http://validator.annodex.org/ :-)
Jack: because we use half-open intervals
michael (on blackboard)
<jackjansen> erik, I wanted to say "because we use half-open intervals we can also do some automatic testing on video/audio"
two test cases: URI resolving & Generate HTTP request (also see photo)
from HTTP request to HTTP response
back on UA:
comment from jack: this is testing from User Expectation ... shouldn't we just test our Spec?
<Yves> we need to test if it's feasible to get interoperable implementations
<Yves> so we test the spec, but also (parts of) implementations as well
<Yves> we won't test that implementations are 100% correct
<Yves> we just test 100% coverage of the spec by interoperable implementations
michael: try the top-down approach (opposed to Jack's proposal of doing it bottom-up)
raphael: in the end both should come up with the same results though
(human in the loop) on UA
<conrad> on the last step ("human in the loop") we should document the expected behaviour of a UA: that the first frame rendered is the first frame of the interval, that the video image is not broken (ie. the decoder has been primed with the previous i-frame)
TestCase 0: URI#
-> == URI == entire resource
Jack: at very least for every dimension we should define all the test cases where we get back the whole resource & also the ones where we get back an empty resource
<guillaume> see http://www.w3.org/2006/07/SWD/RDFa/testsuite/ for example
Yves: there are tools online that check HTTP responses
<Yves> I agree we just need to test that the headers we cared for are ok
michael: we should test our extra HTTP-headers for example ... not the complete HTTP stack
raphael: practically ... just test specific HTTP-headers with name & their possible values
Jack & Conrad: tests are most easily done on server
<guillaume> see http://wiki.csswg.org/test/harness
conrad: two test harnesses ... one for client & one for server
conrad & jack: both test harnesses can be scripted
<jackjansen> I can be scripted, but the qeustion is: do I do what you expect.....
<scribe> ACTION: michael to setup template for test cases on wiki [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action04]
<trackbot> Created ACTION-72 - Setup template for test cases on wiki [on Michael Hausenblas - due 2009-04-24].
raphael: initial 10 test cases ... test case 0 & empty/entire resource & only temporal and npt-units
<conrad> if we are testing a client, the base uri is the uri of the harness
<conrad> if we are testing a server, we need to provide media and test fragment requests to that media
jack: why not only within test harness just the fragment (not the whole URI) to make it automatically testable
erik: what about RTSP?
jack: we do end-to-end test for multiple protocols
raphael: how was the test suite handled within SMIL?
Jack: SMIL 2.0 test suite was xml
& then parsed by bits of perl/python
... then put into HTML forms, again some script processing & afterwards put into a document ... all testing was done manually :(
... please, choose the right media to run the tests!
<raphael> invite zakim #mediafrag
<raphael> trackbot, start telecon
<trackbot> Meeting: Media Fragments Working Group Teleconference
<trackbot> Date: 17 April 2009
<raphael> scribenick: raphael
Wrapping up the test cases
Conrad: have a test case with a
... on a non-media resource, if the behaviour is as expected
Dave: test whether the video.mp4#t=10 (as used by Google video) still works as expected
Michael: we should completement with more negative test cases
Issue 3: Michael, http://www.w3.org/2008/WebVideo/Fragments/tracker/issues/3
Michael: in RFC3986, the
semantics of the fragment is per design not defined
... together with the mime-type, one can learn what the semantics means
... example, HTML, with its registration
... but this is not true for most of the media types
... to register at IANA and IETF the media fragment syntax
Jack: this is not possible, mp4
has already its post-hash syntax
... we could write a cover letter, explaining who we are, and recommend to the mime-type owners what they should register
Michael: there are plenty, is
this possible at all?
... it will take too much time
Member only link: http://www.w3.org/2009/02/06-swcg-minutes.html#item03
<mhausenblas> proposal: media/fragment
Conrad: it's a hack to use the
Accept header for that
... we need to do a media specific documentation
Davy: problem, MPEG-21 has done that for MPEG resources, why they would change and adopt our scheme?
Conrad: we have no authority, we could just convince them with existing implementations, and deployment
Michael: I need to take my action
seriously and come back with figures
... how many mime type owners we need to talk with ?
... how many mime-types will be affected?
Jack: looking at IANA database, it is very hard to guess who is the owner
Michael: my method, look at all
the mime-type concerned (audio/*, video/*, image/*) and look
for each whether they have a semantics for the hash
... if NOT, no problem
... if YES, then, identify the owner and evaluate the impact
... evaluate impact meaning detecting clash
Conrad: we should talk to the AVT
group in IETF
... charter: http://www.ietf.org/html.charters/avt-charter.html
... group home page: http://tools.ietf.org/wg/avt/
... I think Dave Singer is somehow associated to this group
Dave: can you confirm the asumption of Conrad just above?
<dsinger> I am looking for the assumption
Are you associated to some extent to the AVT group in IETF ?
Raphael: I feel the Issue 4 is mixing two things
a) the fact of selecting various tracks, or all tracks except 1
b) pre-selecting track names
Jack: for a) I think it should go
to the version 2 of Media Fragments
... because we will end up soon to a presentation language behind the hash
... the problem is we cannot measure what amount of work it represents, nor to the people we need to talk too
Davy: why not doing like RTSP ...
Raphael: this is the ROE way
Jack: we assume there is a mechanism for which the client can have a description of the media items
Conrad: clarification, the retrieval of the description is not in the media fragments, it is done before
<Yves> so it includes lots of RFCs (not easy to republish)
<mhausenblas> Michael: re issue 3, I mean it is always good to write the cover letter
<mhausenblas> ... just a matter of how the odds are that we can convince them, looking at the sheer number of potential clashes
<mhausenblas> ... where a clash is defined as a media type that defines frag semantics
Yves, hopefully, there will be 0 or 1 RFC impacted
scribe: we really care when there is a clash
<mhausenblas> ... and these semantics contradict with ours
Jack: no, the syntax will not
... we pass verbatim the track="value" to the server, and this is up to the server, capabilities of container formats
Conrad: I agree with you
... I think what we can get is just no audio or no video
Jack: I dont know what get 'audio' means in the general case
<scribe> ACTION: Conrad to change the phrasing of the issue 4 (just audio/video)? [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action05]
<trackbot> Created ACTION-73 - Change the phrasing of the issue 4 (just audio/video)? [on Conrad Parker - due 2009-04-24].
Jack: the issue comes during the implementation trial, where I happily do some transcoding
<Yves> my point was... "it is fuzzy"
<mhausenblas> +q to ask for sanity check
<Yves> I can point to... http://tools.ietf.org/html/draft-ietf-httpbis-p6-cache-06#section-3.6
Raphael: yes, thus the action-62, http://www.w3.org/2008/WebVideo/Fragments/tracker/actions/62 now associated
<Yves> Warning: 214 Transformation applied
<Yves> (note that is is also in rfc2616)
<conrad> 4.3.2 Receipt of Warning: 214 Transformation Applied
<conrad> If the response includes a Warning: 214 Transformation Applied HTTP header, proxies must not apply further transformation.
Proposal from Yves: not really implemented, but we could use the warning header to warn the UA there has been some transcoding
<Yves> oh proxies can do further transformations
<Zakim> mhausenblas, you wanted to ask for sanity check
<Yves> it just warns the client that there was a transformation somewhere in the chain
<conrad> i was quoting from http://www.w3.org/TR/ct-guidelines/
<mhausenblas> Michael: is this within our scope?
<mhausenblas> ... would like to see WG saying: YES!
Jack: should we care, as a standardisation group, whether there is transcoding or not
<conrad> for formats that do not encode presentation in-points, we could add an HTTP response header to inform the client of the presentation time at which to enable rendering
Jack: this is just a heads up,
... we must remember if the server has done some clipping, it must send also the original w/h to the client
... in the case where the spatial cropping has been asked in terms of %
Raphael: my concern, are we
stretching too much the definition of a fragment, when the
fragment (= spatial clipping) has no single bytes in common
from the original resource
... I hope this will be answered by Yve's action to TAG
... perhaps, we will say, this is not anymore a fragment, and recommend to use the ? rather than the # for this use case
Frank: this is an analogy of what is said for spatial but not said for temporal
<davy> scribenick: davy
raphael on the blackboard
raphael: uc & req doc will be
... use cases and requirements and side conditions will be in a separate document
... technologies survey in another document
guillaume: uc & req should be first, then side conditions
raphael: historically, side conditions were first
mhausenblas: put survey in the uc & req doc
proposal: uc & req doc and the technologies survey in one doc?
guillaume: duplicate side conditions (both in working note and REC docs)
jackjansen: in the rec, side conditions will be specified more formally
raphael: spec will be rec
<Yves> wg note as well for UC&req&survey
raphael: uc & req will be working note?
all: working note
<mhausenblas> should http://www.w3.org/2005/Incubator/app-backplane/ be in
<scribe> ACTION: raphael and erik to request feedback of other groups such as SYMM, SVG, HTML + WHATWG, WAI, MAWG, TAG, MobileWeb, TimedText, ... [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action06]
<trackbot> Sorry, couldn't find user - raphael
<scribe> ACTION: Erik and Raphael to request feedback of other groups such as SYMM, SVG, HTML + WHATWG, WAI, MAWG, TAG, MobileWeb, TimedText [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action07]
<trackbot> Created ACTION-74 - And Raphael to request feedback of other groups such as SYMM, SVG, HTML + WHATWG, WAI, MAWG, TAG, MobileWeb, TimedText [on Erik Mannens - due 2009-04-24].
raphael: planning for the next 6
... second public wd within 2-3 months
jackjansen: 2nd wd in september
raphael: if people do not provide
feedback, we should find out why
... Goal is to have a 2nd WD early in september
erik: will the document already be split?
raphael: yes, what about test cases?
mhausenblas: would leave it in the wiki for a while
raphael: current policy regarding
... which policy should we have?
jackjansen: list everyone in the group or list everyone contributed a chapter
mhausenblas: if someone writes
something in the rec, then he/she is an editor
... have two editors, then the contributers
jackjansen: then we can assume that everyone is a contributor
<Yves> yes, all the WG members at least
raphael: looking at other w3c docs, they use editors and a number of contributors
<raphael> Raphael: proposal on the blackboard is to have a number of editors (chairs ? team contact? others?) and all group members as contributors
<scribe> ACTION: raphael to split the document and setup the diff feature of xmlspec [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action08]
<trackbot> Sorry, couldn't find user - raphael
<raphael> trackbot, status?
<scribe> ACTION: raphaël to split the document and setup the diff feature of xmlspec [recorded in http://www.w3.org/2009/04/17-mediafrag-minutes.html#action09]
<trackbot> Created ACTION-75 - Split the document and setup the diff feature of xmlspec [on Raphaël Troncy - due 2009-04-24].
<raphael> BIG THANKS for the local organizers
This is scribe.perl Revision: 1.135 of Date: 2009/03/02 03:52:20 Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: RRSAgent_Text_Format (score 1.00) WARNING: Bad s/// command: s/http://respighi.elis.ugent.be/NinSunaWeb/NN WARNING: Bad s/// command: s/http:\/\/respighi.elis.ugent.be\/NinSunaWeb/NN/g Succeeded: s/aprse/parse/ Succeeded: s/ohw/how/ Succeeded: s/gree/agree/ Succeeded: s/: can be both/: both test harnesses can be/ Succeeded: s/then put into/... then put into/ Succeeded: s/req docs/REC docs/ Found Scribe: Raphael Inferring ScribeNick: raphael Found ScribeNick: raphael Found ScribeNick: mhausenblas Found ScribeNick: erik Found ScribeNick: raphael Found ScribeNick: davy ScribeNicks: raphael, mhausenblas, erik, davy WARNING: Replacing list of attendees. Old list: Mediafrag Silvia Yves New list: Mediafrag Yves Default Present: Mediafrag, Yves Present: Erik Davy Guillaume Michael Jack Conrad Raphael Frank_(observer) Yves_(remote) Silvia_(remote) Dave Agenda: http://www.w3.org/2008/WebVideo/Fragments/wiki/ThirdF2FAgenda Found Date: 17 Apr 2009 Guessing minutes URL: http://www.w3.org/2009/04/17-mediafrag-minutes.html People with action items: conrad erik jack l michael rapha raphael[End of scribe.perl diagnostic output]