Transcript Mining Data to Extract Semantic Concepts

.................................................
MUSCLE e-Team:
Content Analysis Showcase
(CAS)
.................................................
CAS Motivation
.................................................
 We have
– a series of state-of-the-art reports
– a series of presentations by MUSCLE research partners
– a set of specific benchmark data for objective evaluation of
algorithms on specific performance criteria
 Think we need, in addition
– a common understanding what we can potentially do together
by combining our approaches
– a common understanding of what our algorithms are doing
– a joint "practical SoA" showcase
.................................................
CAS Goals
.................................................
 Goals, very briefly and informally:
– to take a common set of multimedia data
– to simply apply whatever algorithms we have on it
(everybody)
– to show others what the results are
– to allow others to use these resuls to build on to of it
 Goals are not:
– benchmarks to evaluate the quantitative performance of
algorithms
– new research in itself, at least not in the initial stages of the
eTeam (may evovle by combining indiv. results)
.................................................
CAS Workplan (1/2)
.................................................
 Some partners record short video sequences:
–
–
–
–
preferably from public TV
in different languages
in an agreed default (low?) quality setting
should comprise a heterogeneous set of characteristics:
• speech, music, noise, persons, animals, objects, fire, ...
– e.g. music video clips, sports, news, soap-opera, commercials,...
 video clips are combined to form a short collective
testbed video used by all partners
 probably made available in different forms
– video
– audio stream
– set of images
.................................................
CAS Workplan (2/2)
.................................................
 All partners apply their algorithms to the joint data
– no need for optimization
– no performance evaluation in terms of quality against benchmark
– "show what you can do"
 Results will be presented
– series of presentations on results and what they look like
 Result data will be shared
–
–
–
–
–
set of feature vectors
set of keyframes
set of detected objects
set of annotated scenes
...
 Phase 2: build upon the available data:
– If I have xxx I can do yyy with my algorithm
.................................................
CAS Timeframe
.................................................
 Months 1-2: data acquisition
 Months 3-7: data analysis
 Month 7: results presentation and joint showcase
 Months 8-12: Combination of data sets
specifically: a range of research exchanges
.................................................
CAS Partners
.................................................
 Vienna University of Technology:
Andreas Rauber, Thomas Lidy, Robert Neumayer
 University of Amsterdam:
Cees Snoek, Nicu Sebe
 Cambridge University:
Julien Fauqueur, Ryan Anderson, Nick Kingsbury
 any many more? everybody?
.................................................
CAS Summary
.................................................
 CAS is a
– Low effort e-Team
– Low research profile e-Team
–
–
–
–
Highly important e-Team to foster collaboration
Highly important e-Team to understand our work
Highly important e-Team to demonstrate our work
Highly important e-Team to unleash potential in MUSCLE
– Network e-Team bringing together what is already there
– Pratical SoA report/showcase
– an e-Team where hopefully many will want to and are able to
participate
.................................................