WG 1
Warning: your browser seems to be quite old, or at least old enough that it does not support the minimal standard capabilities of today's browsers, which are used extensively on this site. Please upgrade to a browser that supports web standards, specifically one that supports minimal CSS1 capabilities, such as Netscape 6 or higher, IE 5 or higher, or Opera 5 or higher. Your browsing experience will be much better! Thanks.
Title Common testing data and frameworks.
Starting date October 2004 (start of Action).
Ending date October 2005.
Coordinator Prof. P. Lobato Correia (IST)
Main activities
Working Group 1 was set up to establish the grounds for effective cooperation along the action duration. As such it started at the beginning of the action and lasted for the first year. The main activities involved:
- Preparation of new test sets to be shared among all COST 292 members, and to be used for common testing purposes;
- Reaching consensus on the usage of common, or at least similar, software platforms to enable effective algorithm sharing and results comparison. This second point was conducted in cooperation with WG2.
Status
This Working Group successfully concluded its activities on October 2005.
After the inaugural meeting in Brussels, where the leader of WG1 was proposed and approved, a mailing list for discussing WG1 issues was setup ([email protected]) and all partners were invited to join the work. The first activity started was the preparation of common test sets – for this purpose a template document to describe the contents to be shared by the action members was prepared.
WG1 met in London (November 2004) and in Riga (June 2005). By then, the test set was more or less stabilized, and the question of the software platforms was the main pending issue, with 3 candidate platforms being under discussion: MUVIS (developed by TUT), SCHEMA (implemented by ITI) and BUSMAN (represented by QMUL and TID). It was then agreed to schedule a meeting specifically to discuss this issue in Lisboa (September 2005), gathering WG1 and WG2 participants.
In the two day meeting in Lisboa, after a presentation of each of the three candidate systems, a mapping of the main objectives of the COST 292 and the potential of each of the possible COST 292 frameworks was done, resulting in a table showing the strong and weak points of each system in view of the various COST 292 goals. It was agreed that the three systems are complementary to each other and that parts of each should be used for different tasks. For instance, MUVIS can be taken as a baseline framework for COST 292, namely to be used for feature extraction purposes, using an API to be extended during the project duration, SCHEMA algorithms may be used for shot boundary detection or segmentation purposes, eventually after adequation to the TRECVID XML format, and BUSMAN relevance feedback module should be further investigated.
The Working Group was officially closed in the Belgrade meeting (November 2005).
Results
A new set of video contents has been made available to the project participants (available at https://cost292.org/), with contributions from:
- University of Bristol/Granada TV:
- 13 sequences of wildlife footage, some manually annotated.
- Sequences for COST 292 internal usage + public dissemination for research and demos.
- VICOM Tech/Basque TV:
- 13 sequences of sports (cycling, pelota) and news. Possibility to make further contents available.
- Two versions for some sequences: with our without the commercial breaks.
- Sequences for COST 292 internal usage + dissemination for research/demos.
- LABRI
- Corpus of video surveillance sequences available on request.
- MUVIS (presented by TUT) – This system was taken as a basis for common software development. An API is available for plugging in new feature extraction modules and the software can be extended to support additional requisites by the COST 292 project.
- BUSMAN (presented by TID/QMUL) – Mainly the relevance feedback mechanisms can be of interest to COST 292.
- SCHEMA (presented by ITI/NTUA) – Mainly the shot boundary and segmentation modules can be of interested for COST 292.