SlideShare a Scribd company logo
1 of 30
Semantic Multimedia Remixing
MediaEval 2013
Search and Hyperlinking Task

Mathilde Sahuguet & Benoit Huet
MediaMixer Technology Framework
Annotation
tools

Analysis
tools

Copyright
tools

©

New Media
Applications
Re-use
Acquire

Fragment
Creation

Fragment
Description

Fragment
Rights

Search

RDF metadata model

Media Assets

Media Fragment URI specification

Storage
(Media +
Metadata)
© 2013 MediaMixer Consortium
17/01/2014 -

http://community.mediamixer.eu

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p2
MediaMixer Technology Framework
Annotation
tools

Analysis
tools

Copyright
tools

©

New Media
Applications
Re-use
Acquire

Fragment
Creation

Fragment
Description

Fragment
Rights

Search

RDF metadata model

Media Assets

Media Fragment URI specification

Storage
(Media +
Metadata)
© 2013 MediaMixer Consortium
17/01/2014 -

http://community.mediamixer.eu

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p3
Summary

 MediaEval Task presentation
 How to Search for Media Fragments
 Examples
 Results
 Conclusion

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p4
Task presentation
 Information seeking in a video dataset:
retrieving video/media fragments

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p5
Task presentation
 Information seeking in a video dataset:
retrieving video/media fragments
 The dataset: 2323 BBC videos of different
genres
 ~1697h of video + audio
 Two types of ASR transcript
 Manual subtitle
 Metadata
 Shot boundaries and keyframes
 Face detection and similarity information
 Concept detection
17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p6
Task presentation
 Search: find a known segment in the collection given a query (text)
<top>
<itemId>item_18</itemId>
<queryText>What does a ball look like when it hits the wall during
Squash</queryText>
<visualCues>ball hitting a wall in slow motion</visualCues>
</top>

 Hyperlinking: find relevant segments relatively to an “anchor”
segment (+- context)
<anchor>
<anchorId>anchor_1</anchorId>
<startTime>13.07</startTime>
<endTime>13.22</endTime>
<item>
<fileName>v20080511_203000_bbcthree_little_britain</fileName>
<startTime>13.07</startTime>
<endTime>14.03</endTime>
</item>
</anchor>
17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p7
Task presentation
 Queries are user generated for both search and
hyperlinking
 Search: 50 queries from 29 users
 Hyperlinking: 98 anchors initially.

 Evaluation:
 For search, searched segments are pre-defined
 For hyperlinking, crowd-sourcing (on 30 anchors only)

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p8
Evaluation measures
 Search
 Mean Reciprocal Rank (MRR): assesses the rank of the
relevant segment



17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

-p9
Evaluation measures
 Search
 Mean Reciprocal Rank (MRR): assesses the rank of the
relevant segment
 Mean Generalized Average Precision (mGAP): takes into
account starting time of the segment
 Mean Average Segment Precision (MASP): measures
both ranking and segmentation of relevant segments



Hyperlinking
 Precision at rank n: how many relevant segment appear in
the top n results
 Mean Average Precision (MAP)

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 10
Our approach – 1. Pre-processing
 Extracting as much information as possible

 Processing ~ 1697h of BBC video data
Visual Concept detection (151)

CERTH

20 days on 100 cores

Scene segmentation

CERTH

2 days on 6 cores

OCR

Fraunhofer

1 day on 10 cores

Keywords extraction

Fraunhofer/UEP 5 hours

Named Entities extraction

Eurecom/UEP

4 days

Face detection and tracking

Eurecom

4 days on 160 cores

1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET
17/01/2014 -

- p 11
Our approach – 2. Indexing
 Solr engine (Lucene) for indexing the data
 Index at different temporal granularities (shot, scene,
sliding window segment)
 Index different features at each temporal granularity
(metadata, ocr, transcripts, visual concepts)

 All information stored in a unified structured
way
 flexible tool to perform the search and hyperlinking

 Next step: design a query for each sub-task

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 12
Solr indexing
 Schema = structure of document using fields of
different types
 Design a schema = decide how to represent
and index the data
<field
<field
<field
<field
<field

name="id" type="string" indexed="true" stored="true" multiValued="false“ required="true" />
name="begin" type="string" indexed="false" stored="true” multiValued="false" required="true"/>
name="end" type="string" indexed="false" stored="true" multiValued="false“ required="true"/>
name="videoId" type="string" indexed="true" stored="true" multiValued="false" required="true"/>
name="subtitle" type="text_en" indexed=“true" stored="true" multiValued="false" required="true"/>

<field name="Actor" type="float" indexed="true" stored="true" multiValued="false" required="true"/>
<field name="Adult" type="float" indexed="true" stored="true" multiValued="false" required="true"/>
<field name=“Animal" type="float" indexed="true" stored="true" multiValued="false" required="true"/>
<etc>

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 13
Solr indexing
 Schema = structure of document using fields of
different types
 Design a schema = decide how to represent
and index the data
<doc>
<field
<field
<field
<field
<field

name="id"> 20080401_013000_bbcfour_legends_marty_feldman_six_degrees_of#t=399,402</field>
name="begin">00:06:39.644</field>
name="end">00:06:42.285</field>
name="videoId">20080401_013000_bbcfour_legends_marty_feldman_six_degrees_of</field>
name="subtitle">'It was very, very successful.'</field>

<field name="Actor">0.143</field>
<field name="Adult">0.239</field>
<field name="Animal">0.0572</field>
</doc>

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 14
Solr indexing
 Analysis step:
 Dependent on each type
 Automatically performed: tokenization, removing stop
words, etc
 It creates tokens that are added to the index
 inverted index
 query is made on tokens

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 15
Our approach – 3. Search
 Using the provided visual cues
 Text search is straighforward
 Visual features? Starting point = visual cues in text form

 Mapping “visual cues” to concepts
 Based on Word-net distances
 Outputs concepts with a confidence score

 Concepts detector confidence
 Manual evaluation of the 100 top images for each concept
=> valid detection rate
17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 16
Example: mapping
 <visualCues>House memories Farm exploration A
poem on animal and shells </visualCues>
Church

Memories
keywords

School
Flags

Farm

Building

Exploration

WordNet
mapping

Dogs
Animal

Poem

Birds

Animal

visual concepts

House

Insect

Shells

Mountain
Cattle

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 17
Example: detector confidence

 100 top images for the concept “Animal”
 58 out of 100 are manually evaluated as valid
17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 18
Our approach – 3. Search
 2 Media Fragment Creation Strategies
 Using pre-constructed segments (scenes)
 Merging segments on the fly (clustering cl10)

 Submitted runs:
 scenes-noC (-C): scenes search using textual (and
visual) features
 scenes-S(-U,-I): scenes search using only textual
features from transcript (S: subtitle, I and U: transcript
type)
 cl10-noC (-C) : Temporal shot clustering within a video
using textual (and visual cues).
17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 19
Our approach – 3. Search
 Pre-filtering (At Video Level)
 Query on the video index: accurate video lies within the
40 top results (36 in the worst case)
 Average position (50 queries) of the correct video: 4,72
 For 23 queries, the 1st video is correct
 For 47 queries, the correct video is within the first 20
videos

 Restricting the media fragment search to the
top videos is likely to increase performances
 identify the optimal number of videos to search

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 20
Solr querying
 Very easy with web interface

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 21
Solr querying
 Very easy with web interface
 Query can be made through http request
 http://localhost:8983/solr/collection_mediaEval/select?q=
text:(Children out on poetry trip Exploration of poetry by
school children Poem writing) Animal:[0.2 TO 1]
Building:[0.2 TO 1]

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 22
Our approach – 4. Hyperlinking

 Reusing the search component
 Designing a query from the anchor
 aligning subtitles to anchor to extract text features
 spotting concepts

 Querying using both scenes and shot clustering
approaches

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 23
Examples



Text query : what to cook with everyday ingredients on a budget, denise
van outen, john barrowman, ainsley harriot, seabass, asparagus,ostrich,
mushrooms, sweet potato, mango, tomatoes
Visual query: denise van outen, john barrowman, ainsley harriot, seabass,
asparagus,ostrich, mushrooms, sweet potato, mango, tomatoes

Expected
20080506_153000_bbctwo_ready_steady_cook.webm#t=67,321

Scenes
20080506_153000_bbctwo_ready_steady_cook.webm#t=48,323

cl10
20080506_153000_bbctwo_ready_steady_cook.webm#t=1287,1406

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 24
Results
 Search

Run

MRR

mGAP

MASP

scenes-C

0.324931

0.187194

0.199647

scenes-noC

0.324603

0.186916

0.199237

scenes-S

0.338594

0.182194

0.210934

scenes-I

0.261996

0.144708

0.158552

scenes-U

0.268045

0.152094

0.164817

cl10-C

0.294770

0.154178

0.181982

cl10-noC

0.286806

0.149530

0.171888

 Scenes give the best performances
 Impact of transcript type: as expected
 Impact of visual concept: smaller than expected
17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 25
mGAP results (60s window)

17/01/2014

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 26
Results
 Hyperlinking
Run

MAP

P-5

P-10

P-20

LA cl10

0.0337

0.3467

0.2533

0.1517

LA scenes

0.1196

0.6133

0.5133

0.3400

LC cl10

0.0550

0.4600

0.4000

0.2167

LC scenes

0.1654

0.6933

0.6367

0.4333

 Scenes offer the best results
 Using context increases performances
 Precision at rank n decreases with n
17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 27
P-10 results (60s windows)

17/01/2014

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 28
Conclusion and Future work
 Searching and HyperLinking Media Fragment
 Scenes perform the best
 Segmentation can still be improved
 Actual algorithm based on visual features
 Future work: including semantics and other features

 Visual features provide limited gain
 Why? Further research needed
 Visual concept detectors accuracy
 Mapping between semantic and visual concepts

17/01/2014

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 29
Related Publications


Sahuguet, M.; Huet, B.; Cervenkova, B.; Apostolidis, E.; Mezaris, V.; Stein, D.;
Eickeler, S.; Redondo Garcia, J. L. & Pikora, L., LinkedTV at MediaEval 2013
search and hyperlinking task, MEDIAEVAL 2013, Multimedia Benchmark
Workshop, October 18-19, 2013, Barcelona, Spain



Stein, D.; Öktem, A.; Apostolidis, E.; Mezaris, V.; Redondo García, J. L.; Troncy,
R.; Sahuguet, M. & Huet, B., From raw data to semantically enriched
hyperlinking: Recent advances in the LinkedTV analysis workflow, NEM
Summit 2013, Networked & Electronic Media, 28-30 October 2013, Nantes,
France



M. Sahuguet and B. Huet. Mining the Web for Multimedia-based Enriching.
Multimedia Modeling MMM 2014, 20th International Conference on MultiMedia
Modeling, 8-10th January 2014, Dublin, Ireland



http://www.multimediaeval.org/mediaeval2013/hyper2013/



http://mediamixer.eu/



http://www.linkedtv.eu/



http://lucene.apache.org/solr/

17/01/2014 -

1st Winter School on Multimedia Processing and Applications
Semantic Multimedia Remixing - B. HUET

- p 30

More Related Content

Similar to Semantic Multimedia Remixing - MediaEval 2013 Search and Hyperlinking Task

Cse373 multimedia systems and design
Cse373    multimedia systems and designCse373    multimedia systems and design
Cse373 multimedia systems and design
Sumit Kasaudhan
 
Handy P@rking Overview
Handy P@rking OverviewHandy P@rking Overview
Handy P@rking Overview
handyparking
 
Sana_Final_Project_Documentation
Sana_Final_Project_DocumentationSana_Final_Project_Documentation
Sana_Final_Project_Documentation
Sameer Srinivas
 
Information and Instructional Technologies
Information and Instructional TechnologiesInformation and Instructional Technologies
Information and Instructional Technologies
Alaa Sadik
 

Similar to Semantic Multimedia Remixing - MediaEval 2013 Search and Hyperlinking Task (20)

A Linked Data Recommender System using a Neighborhood-based Graph Kernel
A Linked Data Recommender System using a Neighborhood-based Graph KernelA Linked Data Recommender System using a Neighborhood-based Graph Kernel
A Linked Data Recommender System using a Neighborhood-based Graph Kernel
 
Web Search - Lecture 10 - Web Information Systems (4011474FNR)
Web Search - Lecture 10 - Web Information Systems (4011474FNR)Web Search - Lecture 10 - Web Information Systems (4011474FNR)
Web Search - Lecture 10 - Web Information Systems (4011474FNR)
 
Mini Project- Personal Multimedia Portfolio
Mini Project- Personal Multimedia PortfolioMini Project- Personal Multimedia Portfolio
Mini Project- Personal Multimedia Portfolio
 
Furuyama - analysis of factors that affect productivity
Furuyama - analysis of factors that affect productivityFuruyama - analysis of factors that affect productivity
Furuyama - analysis of factors that affect productivity
 
Remixing Media on the Semantic Web (ISWC 2014 Tutorial) Pt 1 Media Fragment S...
Remixing Media on the Semantic Web (ISWC 2014 Tutorial) Pt 1 Media Fragment S...Remixing Media on the Semantic Web (ISWC 2014 Tutorial) Pt 1 Media Fragment S...
Remixing Media on the Semantic Web (ISWC 2014 Tutorial) Pt 1 Media Fragment S...
 
Cse373 multimedia systems and design
Cse373    multimedia systems and designCse373    multimedia systems and design
Cse373 multimedia systems and design
 
IRJET- Monument Informatica Application using AR
IRJET-  	  Monument Informatica Application using ARIRJET-  	  Monument Informatica Application using AR
IRJET- Monument Informatica Application using AR
 
Semantic multimedia remixing
Semantic multimedia remixingSemantic multimedia remixing
Semantic multimedia remixing
 
Participatory Media Literacy Uwi2009
Participatory Media Literacy Uwi2009Participatory Media Literacy Uwi2009
Participatory Media Literacy Uwi2009
 
Gareth Jones 1209
Gareth Jones 1209Gareth Jones 1209
Gareth Jones 1209
 
IEEE 2014 ASP.NET with C# Projects
IEEE 2014 ASP.NET with C# ProjectsIEEE 2014 ASP.NET with C# Projects
IEEE 2014 ASP.NET with C# Projects
 
IEEE 2014 ASP.NET with C# Projects
IEEE 2014 ASP.NET with C# ProjectsIEEE 2014 ASP.NET with C# Projects
IEEE 2014 ASP.NET with C# Projects
 
Developing A Semantic Web Application - ISWC 2008 tutorial
Developing A Semantic Web Application -  ISWC 2008 tutorialDeveloping A Semantic Web Application -  ISWC 2008 tutorial
Developing A Semantic Web Application - ISWC 2008 tutorial
 
IRJET- Implementation of Emotion based Music Recommendation System using SVM ...
IRJET- Implementation of Emotion based Music Recommendation System using SVM ...IRJET- Implementation of Emotion based Music Recommendation System using SVM ...
IRJET- Implementation of Emotion based Music Recommendation System using SVM ...
 
LinkedTV @ MediaEval 2013 Search and Hyperlinking Task
LinkedTV @ MediaEval 2013 Search and Hyperlinking TaskLinkedTV @ MediaEval 2013 Search and Hyperlinking Task
LinkedTV @ MediaEval 2013 Search and Hyperlinking Task
 
NJ Wildlife Habitat Finder
NJ Wildlife Habitat FinderNJ Wildlife Habitat Finder
NJ Wildlife Habitat Finder
 
Handy P@rking Overview
Handy P@rking OverviewHandy P@rking Overview
Handy P@rking Overview
 
Sana_Final_Project_Documentation
Sana_Final_Project_DocumentationSana_Final_Project_Documentation
Sana_Final_Project_Documentation
 
Information and Instructional Technologies
Information and Instructional TechnologiesInformation and Instructional Technologies
Information and Instructional Technologies
 
010-chapter-9-multimedia-authoring-packages-revamped1.ppt
010-chapter-9-multimedia-authoring-packages-revamped1.ppt010-chapter-9-multimedia-authoring-packages-revamped1.ppt
010-chapter-9-multimedia-authoring-packages-revamped1.ppt
 

More from MediaMixerCommunity

Opening up audiovisual archives for media professionals and researchers
Opening up audiovisual archives for media professionals and researchersOpening up audiovisual archives for media professionals and researchers
Opening up audiovisual archives for media professionals and researchers
MediaMixerCommunity
 
The Sensor Web - New Opportunities for MediaMixing
The Sensor Web - New Opportunities for MediaMixingThe Sensor Web - New Opportunities for MediaMixing
The Sensor Web - New Opportunities for MediaMixing
MediaMixerCommunity
 
Building a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ ArchivesBuilding a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ Archives
MediaMixerCommunity
 
Media Mixing in the broadcast TV industry
Media Mixing in the broadcast TV industryMedia Mixing in the broadcast TV industry
Media Mixing in the broadcast TV industry
MediaMixerCommunity
 
Building a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ ArchivesBuilding a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ Archives
MediaMixerCommunity
 
Semantic technologies for copyright management
Semantic technologies for copyright managementSemantic technologies for copyright management
Semantic technologies for copyright management
MediaMixerCommunity
 
Intelligent tools-mitja-jermol-2013-bali-7 may2013
Intelligent tools-mitja-jermol-2013-bali-7 may2013Intelligent tools-mitja-jermol-2013-bali-7 may2013
Intelligent tools-mitja-jermol-2013-bali-7 may2013
MediaMixerCommunity
 

More from MediaMixerCommunity (14)

Re-using Media on the Web Tutorial: Introduction and Examples
Re-using Media on the Web Tutorial: Introduction and ExamplesRe-using Media on the Web Tutorial: Introduction and Examples
Re-using Media on the Web Tutorial: Introduction and Examples
 
Opening up audiovisual archives for media professionals and researchers
Opening up audiovisual archives for media professionals and researchersOpening up audiovisual archives for media professionals and researchers
Opening up audiovisual archives for media professionals and researchers
 
The Sensor Web - New Opportunities for MediaMixing
The Sensor Web - New Opportunities for MediaMixingThe Sensor Web - New Opportunities for MediaMixing
The Sensor Web - New Opportunities for MediaMixing
 
Building a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ ArchivesBuilding a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ Archives
 
Media Mixing in the broadcast TV industry
Media Mixing in the broadcast TV industryMedia Mixing in the broadcast TV industry
Media Mixing in the broadcast TV industry
 
Building a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ ArchivesBuilding a linked data based content discovery service for the RTÉ Archives
Building a linked data based content discovery service for the RTÉ Archives
 
Semantic technologies for copyright management
Semantic technologies for copyright managementSemantic technologies for copyright management
Semantic technologies for copyright management
 
Tell me why! ain't nothin' but a mistake describing media item differences w...
Tell me why! ain't nothin' but a mistake  describing media item differences w...Tell me why! ain't nothin' but a mistake  describing media item differences w...
Tell me why! ain't nothin' but a mistake describing media item differences w...
 
A feature analysis based fragment remix instrument
A feature analysis based fragment remix instrumentA feature analysis based fragment remix instrument
A feature analysis based fragment remix instrument
 
Video concept detection by learning from web images
Video concept detection by learning from web imagesVideo concept detection by learning from web images
Video concept detection by learning from web images
 
Fast object re detection and localization in video for spatio-temporal fragme...
Fast object re detection and localization in video for spatio-temporal fragme...Fast object re detection and localization in video for spatio-temporal fragme...
Fast object re detection and localization in video for spatio-temporal fragme...
 
Example-Based Remixing of Multimedia Contents
Example-Based Remixing of Multimedia ContentsExample-Based Remixing of Multimedia Contents
Example-Based Remixing of Multimedia Contents
 
Analysis of visual similarity in news videos with robust and memory efficient...
Analysis of visual similarity in news videos with robust and memory efficient...Analysis of visual similarity in news videos with robust and memory efficient...
Analysis of visual similarity in news videos with robust and memory efficient...
 
Intelligent tools-mitja-jermol-2013-bali-7 may2013
Intelligent tools-mitja-jermol-2013-bali-7 may2013Intelligent tools-mitja-jermol-2013-bali-7 may2013
Intelligent tools-mitja-jermol-2013-bali-7 may2013
 

Recently uploaded

Beyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactBeyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global Impact
PECB
 
1029 - Danh muc Sach Giao Khoa 10 . pdf
1029 -  Danh muc Sach Giao Khoa 10 . pdf1029 -  Danh muc Sach Giao Khoa 10 . pdf
1029 - Danh muc Sach Giao Khoa 10 . pdf
QucHHunhnh
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdf
QucHHunhnh
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
heathfieldcps1
 
Making and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdfMaking and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdf
Chris Hunter
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdf
ciinovamais
 

Recently uploaded (20)

Beyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactBeyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global Impact
 
INDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptx
INDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptxINDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptx
INDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptx
 
1029 - Danh muc Sach Giao Khoa 10 . pdf
1029 -  Danh muc Sach Giao Khoa 10 . pdf1029 -  Danh muc Sach Giao Khoa 10 . pdf
1029 - Danh muc Sach Giao Khoa 10 . pdf
 
How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdf
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
 
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptxBasic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
 
ICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptxICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptx
 
ComPTIA Overview | Comptia Security+ Book SY0-701
ComPTIA Overview | Comptia Security+ Book SY0-701ComPTIA Overview | Comptia Security+ Book SY0-701
ComPTIA Overview | Comptia Security+ Book SY0-701
 
Making and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdfMaking and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdf
 
microwave assisted reaction. General introduction
microwave assisted reaction. General introductionmicrowave assisted reaction. General introduction
microwave assisted reaction. General introduction
 
Measures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeMeasures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and Mode
 
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
 
This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.
 
ICT role in 21st century education and it's challenges.
ICT role in 21st century education and it's challenges.ICT role in 21st century education and it's challenges.
ICT role in 21st century education and it's challenges.
 
Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104
 
Introduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsIntroduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The Basics
 
Unit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxUnit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptx
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdf
 
Web & Social Media Analytics Previous Year Question Paper.pdf
Web & Social Media Analytics Previous Year Question Paper.pdfWeb & Social Media Analytics Previous Year Question Paper.pdf
Web & Social Media Analytics Previous Year Question Paper.pdf
 

Semantic Multimedia Remixing - MediaEval 2013 Search and Hyperlinking Task

  • 1. Semantic Multimedia Remixing MediaEval 2013 Search and Hyperlinking Task Mathilde Sahuguet & Benoit Huet
  • 2. MediaMixer Technology Framework Annotation tools Analysis tools Copyright tools © New Media Applications Re-use Acquire Fragment Creation Fragment Description Fragment Rights Search RDF metadata model Media Assets Media Fragment URI specification Storage (Media + Metadata) © 2013 MediaMixer Consortium 17/01/2014 - http://community.mediamixer.eu 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p2
  • 3. MediaMixer Technology Framework Annotation tools Analysis tools Copyright tools © New Media Applications Re-use Acquire Fragment Creation Fragment Description Fragment Rights Search RDF metadata model Media Assets Media Fragment URI specification Storage (Media + Metadata) © 2013 MediaMixer Consortium 17/01/2014 - http://community.mediamixer.eu 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p3
  • 4. Summary  MediaEval Task presentation  How to Search for Media Fragments  Examples  Results  Conclusion 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p4
  • 5. Task presentation  Information seeking in a video dataset: retrieving video/media fragments 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p5
  • 6. Task presentation  Information seeking in a video dataset: retrieving video/media fragments  The dataset: 2323 BBC videos of different genres  ~1697h of video + audio  Two types of ASR transcript  Manual subtitle  Metadata  Shot boundaries and keyframes  Face detection and similarity information  Concept detection 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p6
  • 7. Task presentation  Search: find a known segment in the collection given a query (text) <top> <itemId>item_18</itemId> <queryText>What does a ball look like when it hits the wall during Squash</queryText> <visualCues>ball hitting a wall in slow motion</visualCues> </top>  Hyperlinking: find relevant segments relatively to an “anchor” segment (+- context) <anchor> <anchorId>anchor_1</anchorId> <startTime>13.07</startTime> <endTime>13.22</endTime> <item> <fileName>v20080511_203000_bbcthree_little_britain</fileName> <startTime>13.07</startTime> <endTime>14.03</endTime> </item> </anchor> 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p7
  • 8. Task presentation  Queries are user generated for both search and hyperlinking  Search: 50 queries from 29 users  Hyperlinking: 98 anchors initially.  Evaluation:  For search, searched segments are pre-defined  For hyperlinking, crowd-sourcing (on 30 anchors only) 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p8
  • 9. Evaluation measures  Search  Mean Reciprocal Rank (MRR): assesses the rank of the relevant segment  17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET -p9
  • 10. Evaluation measures  Search  Mean Reciprocal Rank (MRR): assesses the rank of the relevant segment  Mean Generalized Average Precision (mGAP): takes into account starting time of the segment  Mean Average Segment Precision (MASP): measures both ranking and segmentation of relevant segments  Hyperlinking  Precision at rank n: how many relevant segment appear in the top n results  Mean Average Precision (MAP) 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 10
  • 11. Our approach – 1. Pre-processing  Extracting as much information as possible  Processing ~ 1697h of BBC video data Visual Concept detection (151) CERTH 20 days on 100 cores Scene segmentation CERTH 2 days on 6 cores OCR Fraunhofer 1 day on 10 cores Keywords extraction Fraunhofer/UEP 5 hours Named Entities extraction Eurecom/UEP 4 days Face detection and tracking Eurecom 4 days on 160 cores 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET 17/01/2014 - - p 11
  • 12. Our approach – 2. Indexing  Solr engine (Lucene) for indexing the data  Index at different temporal granularities (shot, scene, sliding window segment)  Index different features at each temporal granularity (metadata, ocr, transcripts, visual concepts)  All information stored in a unified structured way  flexible tool to perform the search and hyperlinking  Next step: design a query for each sub-task 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 12
  • 13. Solr indexing  Schema = structure of document using fields of different types  Design a schema = decide how to represent and index the data <field <field <field <field <field name="id" type="string" indexed="true" stored="true" multiValued="false“ required="true" /> name="begin" type="string" indexed="false" stored="true” multiValued="false" required="true"/> name="end" type="string" indexed="false" stored="true" multiValued="false“ required="true"/> name="videoId" type="string" indexed="true" stored="true" multiValued="false" required="true"/> name="subtitle" type="text_en" indexed=“true" stored="true" multiValued="false" required="true"/> <field name="Actor" type="float" indexed="true" stored="true" multiValued="false" required="true"/> <field name="Adult" type="float" indexed="true" stored="true" multiValued="false" required="true"/> <field name=“Animal" type="float" indexed="true" stored="true" multiValued="false" required="true"/> <etc> 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 13
  • 14. Solr indexing  Schema = structure of document using fields of different types  Design a schema = decide how to represent and index the data <doc> <field <field <field <field <field name="id"> 20080401_013000_bbcfour_legends_marty_feldman_six_degrees_of#t=399,402</field> name="begin">00:06:39.644</field> name="end">00:06:42.285</field> name="videoId">20080401_013000_bbcfour_legends_marty_feldman_six_degrees_of</field> name="subtitle">'It was very, very successful.'</field> <field name="Actor">0.143</field> <field name="Adult">0.239</field> <field name="Animal">0.0572</field> </doc> 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 14
  • 15. Solr indexing  Analysis step:  Dependent on each type  Automatically performed: tokenization, removing stop words, etc  It creates tokens that are added to the index  inverted index  query is made on tokens 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 15
  • 16. Our approach – 3. Search  Using the provided visual cues  Text search is straighforward  Visual features? Starting point = visual cues in text form  Mapping “visual cues” to concepts  Based on Word-net distances  Outputs concepts with a confidence score  Concepts detector confidence  Manual evaluation of the 100 top images for each concept => valid detection rate 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 16
  • 17. Example: mapping  <visualCues>House memories Farm exploration A poem on animal and shells </visualCues> Church Memories keywords School Flags Farm Building Exploration WordNet mapping Dogs Animal Poem Birds Animal visual concepts House Insect Shells Mountain Cattle 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 17
  • 18. Example: detector confidence  100 top images for the concept “Animal”  58 out of 100 are manually evaluated as valid 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 18
  • 19. Our approach – 3. Search  2 Media Fragment Creation Strategies  Using pre-constructed segments (scenes)  Merging segments on the fly (clustering cl10)  Submitted runs:  scenes-noC (-C): scenes search using textual (and visual) features  scenes-S(-U,-I): scenes search using only textual features from transcript (S: subtitle, I and U: transcript type)  cl10-noC (-C) : Temporal shot clustering within a video using textual (and visual cues). 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 19
  • 20. Our approach – 3. Search  Pre-filtering (At Video Level)  Query on the video index: accurate video lies within the 40 top results (36 in the worst case)  Average position (50 queries) of the correct video: 4,72  For 23 queries, the 1st video is correct  For 47 queries, the correct video is within the first 20 videos  Restricting the media fragment search to the top videos is likely to increase performances  identify the optimal number of videos to search 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 20
  • 21. Solr querying  Very easy with web interface 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 21
  • 22. Solr querying  Very easy with web interface  Query can be made through http request  http://localhost:8983/solr/collection_mediaEval/select?q= text:(Children out on poetry trip Exploration of poetry by school children Poem writing) Animal:[0.2 TO 1] Building:[0.2 TO 1] 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 22
  • 23. Our approach – 4. Hyperlinking  Reusing the search component  Designing a query from the anchor  aligning subtitles to anchor to extract text features  spotting concepts  Querying using both scenes and shot clustering approaches 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 23
  • 24. Examples   Text query : what to cook with everyday ingredients on a budget, denise van outen, john barrowman, ainsley harriot, seabass, asparagus,ostrich, mushrooms, sweet potato, mango, tomatoes Visual query: denise van outen, john barrowman, ainsley harriot, seabass, asparagus,ostrich, mushrooms, sweet potato, mango, tomatoes Expected 20080506_153000_bbctwo_ready_steady_cook.webm#t=67,321 Scenes 20080506_153000_bbctwo_ready_steady_cook.webm#t=48,323 cl10 20080506_153000_bbctwo_ready_steady_cook.webm#t=1287,1406 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 24
  • 25. Results  Search Run MRR mGAP MASP scenes-C 0.324931 0.187194 0.199647 scenes-noC 0.324603 0.186916 0.199237 scenes-S 0.338594 0.182194 0.210934 scenes-I 0.261996 0.144708 0.158552 scenes-U 0.268045 0.152094 0.164817 cl10-C 0.294770 0.154178 0.181982 cl10-noC 0.286806 0.149530 0.171888  Scenes give the best performances  Impact of transcript type: as expected  Impact of visual concept: smaller than expected 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 25
  • 26. mGAP results (60s window) 17/01/2014 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 26
  • 27. Results  Hyperlinking Run MAP P-5 P-10 P-20 LA cl10 0.0337 0.3467 0.2533 0.1517 LA scenes 0.1196 0.6133 0.5133 0.3400 LC cl10 0.0550 0.4600 0.4000 0.2167 LC scenes 0.1654 0.6933 0.6367 0.4333  Scenes offer the best results  Using context increases performances  Precision at rank n decreases with n 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 27
  • 28. P-10 results (60s windows) 17/01/2014 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 28
  • 29. Conclusion and Future work  Searching and HyperLinking Media Fragment  Scenes perform the best  Segmentation can still be improved  Actual algorithm based on visual features  Future work: including semantics and other features  Visual features provide limited gain  Why? Further research needed  Visual concept detectors accuracy  Mapping between semantic and visual concepts 17/01/2014 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 29
  • 30. Related Publications  Sahuguet, M.; Huet, B.; Cervenkova, B.; Apostolidis, E.; Mezaris, V.; Stein, D.; Eickeler, S.; Redondo Garcia, J. L. & Pikora, L., LinkedTV at MediaEval 2013 search and hyperlinking task, MEDIAEVAL 2013, Multimedia Benchmark Workshop, October 18-19, 2013, Barcelona, Spain  Stein, D.; Öktem, A.; Apostolidis, E.; Mezaris, V.; Redondo García, J. L.; Troncy, R.; Sahuguet, M. & Huet, B., From raw data to semantically enriched hyperlinking: Recent advances in the LinkedTV analysis workflow, NEM Summit 2013, Networked & Electronic Media, 28-30 October 2013, Nantes, France  M. Sahuguet and B. Huet. Mining the Web for Multimedia-based Enriching. Multimedia Modeling MMM 2014, 20th International Conference on MultiMedia Modeling, 8-10th January 2014, Dublin, Ireland  http://www.multimediaeval.org/mediaeval2013/hyper2013/  http://mediamixer.eu/  http://www.linkedtv.eu/  http://lucene.apache.org/solr/ 17/01/2014 - 1st Winter School on Multimedia Processing and Applications Semantic Multimedia Remixing - B. HUET - p 30