[video-codec] Notes from NETVC BOF

[Cullen Jennings <[e-mail]>]

Chairs - Jon Peterson, Adam Roach

Note takers - Varum Singh, Cullen Jennings 

AD Set Context 
At IETF75, we tried to decide if we should do an audio codec. That resulted in OPUS an widely successful audio codec. Later we decided to start looking at video codec. Today we are trying to decide what we are going to a video codec WG. 


Chairs
--------

- see slide for clear set of goals of what we want to accomplish 

AD & Chairs clarified purpose of today was to decide if we should do something in this space and what it should be 



Key Consideration presentation by Mo
--------------------------------------------------

Mo discussed several of the issues that are bit different for an codec meant for the Internet. 

Discussed how our need for fast fine rate control is a different set of requirements than what most video codec are optimized for. 

Algorithm Agility 
- brilliant ideas solicited for way to avoid IPR risk and to do that technically to allow algorithm agility to minimize IPR risk 


Question
Q what resolutions and bit depth?
A: Very rough requirements draft mentions this but the WG needs to

Q. Keith - new codecs make new problems (transcoding). Should be in requirements that don't make transcoding more difficult. Concern that we can address the delay problem in transcoding 

Q. Tim T - IPR around spatial scaling is nasty, do you think it is possible to more than? 
A. 

Q. Bernard - in many cases spatial scaling is less efficient than simulcast
Opus had an ease of configuration that made it successful. Do we want in band FEC, are there are other things like this. When doing switching, you send FIR and get huge hit in bandwidth, and spacial support in bitstream might be able to help with that. 

Q Jean-Marc - opus was designed for RTP and we need to do the same thing here 


Tim T presentation on the Daala project 

Strategy for getting RF free
- replace major chunks of the codec with very different technology to avoid IPR 
- this is high risk but high reward
- helps avoid large swaths of patents including ones we are not aware of 
- moves us area that is much less patent s

Explained several areas that are huge parts of the key way most codecs work today and then went on to explain the very different tackiness that daala is exploring to to avoid theses techniques. 

Example given in sides that show an example where image looks better but PSNR is worse. 

Contributions to daala code from 37 people - many different companies - many of them IETF participants

Q - Frank Bosson - Questions about the type of encoding and which 265.
A - Using just one I frame on sequence 1 second long and trying to set up for an interactive type codec. 
Q - what about other codec
A - yes and don't look as good as this but like some other metrics too 
A - Time suggested taking results with full shaker of salt


Q- Mo Z. - External view of this work means that we need to really focus on evaluation. We need metrics to focus on true real time performance. Traditional codecs have fallen way below reference implementation when used in real time practical implementations. 

Q. Eric Rescorla - ask about graph 
A. For 1080p you tube around 6 and low quality video conferencing around 1mbps or 0.5
Q about continuous improvement
A. have web site (https://arewecompressedyet.com/) that shows how compression is going on data sets provided by anyone on many codecs

Q. how fast does the test framework need to be 
A. run of a bunch of video sequences takes about 20 minutes 

Q Frank Bosson - Probably want to be at around 2-3 mbps for 1080p not the 6 mbps listed before. 

Q. Ross Finlayson - has everyone signed the equivalent of Note Well 
A. no 


Charter Discussion
-------------------------

Chairs presented objective and the charter to the room 

Q. Keith D asked about used for interactive or non interactive
A. yes this is for interactive 
A. pointed at point 2 and 3 and processes 1/2 slide
Keith D felt what was in chart was pretty good but might send a few words to slightly more mention real time

Bernard - We should say where it is better. He thinks it has to be better. Opus is a quantum leap better than stuff before it. Would like it to be better than existing stuff in way other that RF. He wants something better like auto configuration to do in band FEC. Some area can say just want to be competitive with, but in some area we want to say we are way better - and not just "freer"

- We should be very generic in how we address this because we are not the 

HTA - disagrees with Bernard. If we start saying things like MUST be better than popular stuff and goal is 2 years out,   . Opus is just a little bit better than AMR-WB. Like charter text as is. 

Richard Barnes - the must be better is not really relevant to charter. Its the consensus of WG to say it is done that is key thing. The IESG is not evaluating if it is better

Bernard A. - In band FEC is something that has not been done. Just need something to 

Keith Drage -  If the only thing we can say about it is a RF codec, then we have failed. RF is just a part of it. Charter should say we want to develop a better codec

Steve Bosco - we might want to add the words "low latency" somewhere to indicate where we are focusing it to be better. Good to a

Gaelle Martin-Cocher - 
Q - compare to codecs when it is done is vague. It should say it needs to compare to something specific, VP9, 265 etc

Q. have you considered allow a core profile that is RF and non core profile that is not RF 

Q. ITU has been trying to do RF codec for awhile now 


Eric Rescorla 
- need to remove a } on some bullet 

- would be nice to have some word around diversity of software like if we want to be able to do software implementation on mobile phones we should say that. Is the assumptions software on 


Thomas Edwards with box 
- might be worth doing gap analysis on existing codecs 
- people are working on jpeg2000 profile that can provide a low latency video codec 
- even with mepg2 we are seeing improvements over time as people learn to use the tools better.
- with H265 we are seeing it is not appropriate for software for all resolutions and frame rates. It is only now that we are getting H265 


Tim Terriberry - responding to Gaelle questions
- we have tried the approach of  RF but not as good - theora for example - not as successful as we want 
- with 264 we had plan to make baseline RF and then other profiles non RF but baseline did not end up RF

Gaelle - We should reach out to other places
Chairs - we try and look and see if we expertise. Same thing with opus, many people doubted this is right place but we took that risk. Downside is we could have RFC that does not light world on fire. 
Gaelle - If you want to mitigate risk of IPR, you will want to have the ITU and may have better change to get a broader set of companies to signal IPR on it. 
Gaelle - how do you know what IRP risk are acceptable 
Chairs - IETF process leaves that the participants of the WG 

Ted Hardie - some of these issues are dealt with on further slides

Mo Z - Might be good to have a list of useful IPR 

Randall Jesup - Better is a rat hole. RF and competitive for interactive / real time is what we need. 

Thomas - There is always the risk that there is IPR you will not know about. 
Chairs - this is true about all the stuff we work on 

Bernard - it will be better to get very low bit rates because of  lower price HTMl phones in Africa. 

Mo - propose strike the 256 kbps text. 
Tim support

**** Chair ask if anyone want to strike the text on 256 kbps lower bow. No one objected. 

Joe - we have general liaison to ITU but do we need specific liaison to to SQ16
**** Chair will follow up offline to see what need

Stephan Wenger - currently liaison  to mpeg. Most currently work is joint work between ITU and MPEG. Stephen 

Bernard - no payload format. 
Answer - we will do that in payload at same time

Ted Hardie - first document are huge rat whole and mostly unless

Mo - be good to have a survey of relevant IPR that is expired or soon expiring 
Chairs - might what as living doc not a deliverable that IETF consensus
evaluation metrics should be evaluation methodology
perhaps we should define encoded bitstream and decoder operation instead of defining encoder and decoder

Frank - right now read as we will normatively define encoder. is that really what we want 

Cullen - explain desire for the point 1 text

Keith - point one on IRP could be read as changing the way this WG will deal IPR 

Ted - will have to have argument about what licensing terms are before they know what they will contribute 
We could have the WG try to say if some of the well known licenses are acceptable to WG. 

Tim -  What ted was proposing is in line with what intended here and could ted send text 

Stephen Wenger - done patent work for decade. When you say you have a goal of RF, people can dream up something. Recommendation is to stop this effort.  Go ahead and do this for open source. 

Harald A. - BCP 79 does not have licensing term. 

Eric Rescorla - don't feel strongly about it. .  One proposal would move it down to item 4. Another possible possibility to rephrase as example terms. 

***** No one felt strongly the IPR number one bullet point should stay in the charter. 

On to milestones slides
- could strike the IPR milestone as well 

Frank - Question about dates - how long are we talking
Tim T - hope to freeze bitstream by end of year. A year is optimistic but that type of range. Opus took 3 years from point is chartering. This is a larger chunk works. More people working on it.


Gaelle - ?
Answer - invite anyone who show up and contribute encoder / decoder 


Keith - take a pass though and check milestones match 


Moving into Questions

The question - Is there a problem worth solving ?


Thomas - ask about what the problem is 

Strong and unanimous for the minutes in favor of YES 


Is the scope of problem well defined and understood? Is there agreement on what a WG would need to deliver?



Strongly consensus with a little bit of decidedly in favor of this




Are there people will do do the work 

7  people willing to writ the documents 

lots of people willing to review drafts 

big chunk of people willing to implement 



How many people feel that a WG should not be formed at the IETF ?

Heard maybe 1 person 

Majority of room strongly in favor of forming a WG.