photo clustering of social events by extending phototoc to a rich context
DESCRIPTION
Presented at ICMR 2014 Workshop on Social Events in Web Multimedia (SEWM). http://mklab2.iti.gr/sewm14/ The popularisation of the storage of photos on the cloud has opened new opportunities and challenges for the organisa- tion and extension of photo collections. This paper presents a light computational solution for the clustering of web pho- tos based on social events. The proposal combines a first over-segmentation of the photo collections of each user based on temporal cues, as previously proposed in PhotoTOC. On a second stage, the resulting mini-clusters are merged based on contextual metadata such as geolocation, keywords and user IDs. Results indicate that, although temporal cues are very relevant for event clustering, robust solutions should also consider all these additional features.TRANSCRIPT
Photo Clustering of Social Events by Extending PhotoTOC to a Rich Context
Daniel Manchón-Vizuete Irene Gris-Sarabia Xavier Giró-i-Nieto
ICMR 2014 Workshop on Social Events in Web Multimedia, Glasgow (Scotland/UK), 01/04/2014
OutlineMotivation Related workApproachResultsFuture workConclusions
Photo Clustering of Social
Events by Extending
PhotoTOC to a Rich Context
D. Manchón-Vizuete I. Gris-Sarabia X. Giró-i-Nieto
Motivation
Motivation
Motivation
Credit: National Institute on Aging and FDA.
Motivation
Motivation
Credit: Alan Smeaton and Cathal Gurrin (Dublin City University, 2013)
OutlineMotivation Related workApproachResultsFuture workConclusions
Photo Clustering of Social
Events by Extending
PhotoTOC to a Rich Context
D. Manchón-Vizuete I. Gris-Sarabia X. Giró-i-Nieto
Related work
PhotoTOC[Platt et al, PACRIM 2003]
Related work
[Becker, Naaman & Gravano, ACM WSDM 2010]
Rich context
User-provided annotations
Automatically generated information
TitleDescription
Creation timeGeo-location
OutlineMotivation Related workApproachResultsFuture workConclusions
Photo Clustering of Social
Events by Extending
PhotoTOC to a Rich Context
D. Manchón-Vizuete I. Gris-Sarabia X. Giró-i-Nieto
Approach(a) Temporal sorting by each user independently
Hi, I’m John. Hi, I’m Emily.
Approach(b) Temporal-based oversegmentation in mini-clusters
PhotoTOC[Platt et al, PacRim 2003]
Approach(b) Temporal-based oversegmentation in mini-clusters
Approach(c) Sequential merging of mini-clusters
? tavg(·) avg(·) avg(·)avg(·)
Approach(c) Sequential merging of mini-clusters
Weightedmodalities
● creation (or upload) time● geolocation● textual labels● same user
Approach(c) Sequential merging of mini-clusters
Geolocation (d=haversine)Time stamp (d=L1)
Text labels (d=Jaccard) Same user (d=boolean)
Approach(c) Sequential merging of mini-clusters
Approach(c) Sequential merging of mini-clusters
Mean and std. deviation learned on
pairs of photos within the same training event.
Approach(c) Sequential merging of mini-clusters
phi function
Approach(c) Sequential merging of mini-clusters
decision threhold
Approach(c) Sequential merging of mini-clusters
OutlineMotivation Related workApproachExperimentsFuture workConclusions
Photo Clustering of Social
Events by Extending
PhotoTOC to a Rich Context
D. Manchón-Vizuete I. Gris-Sarabia X. Giró-i-Nieto
DatasetReSEED - Social Event Detection Dataset
Timo Reuter, Symeon Papadopoulos, Vasilios Mezaris & Philipp Cimiano (ACM MM Sys 2014)[Paper: http://dx.doi.org/10.1145/2557642.2563674] [Dataset: http://greententacle.techfak.uni-bielefeld.de/reseed/]
Results
Results
Results
Diversity of cameras (qualities)
Diversity of locations
Results
Results
Visual redundancy (not exploited)
Results
False split
Results
False merge
OutlineMotivation Related workApproachExperimentsFuture workConclusions
Photo Clustering of Social
Events by Extending
PhotoTOC to a Rich Context
D. Manchón-Vizuete I. Gris-Sarabia X. Giró-i-Nieto
Future work
● Median-based normalisation for robustness against outliers.
● Better optimisation of parameters K and d.
● Improve fusion approach (SVM ?).
● Event-dependent merging criteria.
● Efficient use of visual features (higher computation).
OutlineMotivation Related workApproachExperimentsFuture workConclusions
Photo Clustering of Social
Events by Extending
PhotoTOC to a Rich Context
D. Manchón-Vizuete I. Gris-Sarabia X. Giró-i-Nieto
Conclusions● All considered context (geo, user, tags) can help.
● Watch out for outliers.
● Divide and conquer… sequentially.
Thank you
SEWM !
Thank you !
http://bitsearch.blogspot.com
https://imatge.upc.edu/web/people/xavier-giro
@DocXavi
/ProfessorXavi