Author: Raphael, Zeljko, Lynda
CWI participants: Raphael, Zeljko, Lynda
# participants: around 30
COMM status: only audio descriptors are now missing. One can extend COMM using the procedure in D4.6
The doc of the COMM API is currently generated by Javadoc.
Annotation of 4 min of TrecVID using COMM. The main problem is the performance as the number of triples
generated is about 250 K! The annotations are large due to the D&S approach of the modeling of the
ontology.
Open issues: Inferecing adds many anonymous classes; High memory consumption of COMMAPI; Missing retrieval
functions.
M-OntoMat-Annotizer v2.0 would be available on the ITI web site for download and re-use.
Work continued on Networked graphs.
CWI could report in the M24 deliverable the work on the use of SKOS for converting the NewsCodes.
Three knowledge bases available from the SMART-Web project: player / match and event in soccer. These knowledge bases could be reused in a possible world cup scenario in the news activity.
Presentation of the Ex tool (soon in the Dublin repository) which does web information extraction. Similar to GATE or KIM in objectives. Might be interesting to test it!
KAT develompent SVN + Trac:
https://kater.uni-koblenz.de/svn/montomat2
Read only access: kspace / kspace
A alpha version of KAT has been demonstrated. KU is taking care of releasing version
Long discussion about the plugin architecture which seems to be a good solution: separation of analysis
pluggins versus vizualisation pluggins (for images, for videos, for audio, etc.)
Alan showed the Video Olympics 2007 (trailer video), held with CIVR last July in Amsterdam. There will be a Video Olympics @ CIVR 2008 where K-Space will compete with its interactive search system.
K-Space book: EPFL puts back its book on the table, but officially withdraws it after some (not so) fighty discussion.
SAMT conference series: raise concern about the quality for this year's conference. Clarify what is the role of the SAMT steering board. To be discussed during the TMC.
There will be the SMaRT constitution day meeting the 6th of December in the afternoon. Proposal for making that during the lunch.
SAMT conference series: quality control issue.
SAMT steering board: AP for Craig to create formally this list (Lynda should be back in)
Alan will produce a draft charter explaining the role of the SAMT steering board (rotation of membership,
chairman).
Summer School series: the 3rd one will be organized by ATC (with ITI/NTUA) in Crete, sponsored by MESH and K-Space. How K-Space can keep an eye on the lectures program? SMaRT will take over. Joemon and Gael will be in the steering board of this year organization?
PhD Jamboree: organise another one in 2008, under the SMaRT umbrella (K-Space will sponsor it).
K-Space sequel: next relevant call is 3rd call, deadline for proposal would be April 2008.
Ebroul asked if someone wants to drive a proposal, gathering a subset of some K-Space partners.
The question is whether there will be NoEs for the 3rd call. It seems there is appetite in the room
to make another NoE, of perhaps 10 partners. Ebroul would prefer to not lead such a proposal, but just
get involved. Vojtech will try to gather some bulleted points of a possible proposal.
A telecon could be organized on: Friday 19th, 12:00 AM Paris/Amsterdam.
Alan open the meeting repeating the story from ACM Multimedia. Good news is that the partners want to do the task broader than necessary for TrecVID, in order to apply the summarization framework for other K-Space data. They would like to use it for personal video data summarization, but other topics are also open. Alan also has the idea to try to use the system at the final K-Space review meeting, so that we capture breakfast or early activities that day, and then show 3 min summarization during the review. There were some ideas (also from Alan) that Lynda should get some funding for chocolates, and to bring the chocolates at the final review as that would introduce diversity to final review video, and potentially improve the summarization video.
Some of the low-level features that the partners want to use could potentially be useful for us, such as face tracking, and motion dynamics of the scene. The idea of the partners that will participate in this task is to work iteratively, first listing everything that the partners could offer, and then narrowing down and selecting useful ones, and developing new feature detectors for a concrete task. The plan is to avoid too rigid focusing on BBC summarization task, but also to identify potential use of available resources for other purposes. They will send detailed plan to K-Space list to keep in the loop other partners.
This year the summarization task appeared for the first time at the TrecVID, so it was hard to see compare the systems and approaches. However, there are some interesting observations. First, users like to have nice and smooth animation in the summarization video. DCU, for example, was extracting K-frames (images from the video taken every few seconds), and then showed the selection of these static images, presenting every image a second or two. Users liked much more when these seconds or two where real videos with smooth animation, event in fast forward mode. Second, all the advanced interfaces performed poorly in comparison with pure video summarization. For example, some partners have additional elements, such as story lines, or separate images for redundant parts of the scene, but this seems to introduce too much visual noise. Fast forward element of video was also well accepted by the users.