[rmcat] [draft-ietf-rmcat-video-traffic-model] New video traces for syncodecs

Sergio Mena <semena@cisco.com> Mon, 07 March 2016 19:35 UTC

To: "rmcat@ietf.org" <rmcat@ietf.org>
From: Sergio Mena <semena@cisco.com>
Message-ID: <56DDD80C.1030206@cisco.com>
Date: Mon, 07 Mar 2016 20:35:40 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.5.1
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"; format="flowed"
Content-Transfer-Encoding: 7bit
Archived-At: <http://mailarchive.ietf.org/arch/msg/rmcat/kvXUtCIbvbbgIWZb0ylHCIEH7do>
Cc: "Xiaoqing Zhu (xiaoqzhu)" <xiaoqzhu@cisco.com>
Subject: [rmcat] [draft-ietf-rmcat-video-traffic-model] New video traces for syncodecs
Precedence: list

Hi all,

We would like to announce that a new set of video traces is available as 
part of syncodecs [1]: the C++ implementation of the video traffic 
models described in [2].

The main issue with the old set of video traces was their short length: 
10-12 secs. For this reason, the trace-based codecs tended to show a 
periodic pattern in the delay plots we obtained when using them to test 
rmcat candidate algorithms.

The two new video traces have been generated from raw frames spanning 
over a period of 250 seconds, which is long enough to use them in the 
test cases described in [3].

The original videos from which the traces have been obtained are not 
"talking head style" video conferencing videos. They are short animation 
movies. Unfortunately, after searching the web for days, we haven't been 
able to find a "talking head" video that suits our requirements:

* At least 1080p as resolution
* At least 120 seconds long

The sites we have searched include those that have been announced in 
this mailing list in the past, namely:
* http://www.netlab.tkk.fi/~varun/test_sequences/ (announced by Varun 
Singh in July 2015)
* http://downloads.webmproject.org/ietf_tests/ (announced by Harald 
Alvestrand in August 2015)

We would be very grateful if someone can provide us with a "talking 
head" sequence that fulfils the requirements stated above (res >= 1080p, 
length >=2 mins). If someone comes up with such a sequence (.y4m format 
would be best), we'll be happy to generate the traces and update 
syncodecs accordingly.

For the sake of completeness, here is a high-level description on how we 
generate the traces. We use videolan's x264 library (with resizing 
capabilities), configured with infinite GOP, no scenecut, lookahead 
value of 1, resized to the target resolution, and other less important 
parameters (see the traces tarball in syncodecs for the exact command 
line we use). If someone is interested in further details on how we 
produce the video traces (e.g., in order to do it yourself with your own 
sequences), do not hesitate to get back to me.

Best regards,

Sergio

References:
[1] https://github.com/cisco/syncodecs
[2] draft-ietf-rmcat-video-traffic-model-00
[3] draft-ietf-rmcat-eval-test-02

[rmcat] [draft-ietf-rmcat-video-traffic-model] Ne… Sergio Mena