BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//TUC//Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:Europe/Athens
TZNAME:EEST
DTSTART:19700329T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:EET
DTSTART:19701025T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CREATED:20240110T123139Z
LAST-MODIFIED:20240110T123139Z
DTSTAMP:20260909T121853Z
UID:1788945533@tuc.gr
SUMMARY:Ομιλία Dr. Γεωργίου Κορδοπάτη-Ζήλου 
 "Retrieving same incident videos wit
 h similarity learning"
LOCATION:Λ - Κτίριο Επιστημών/ΗΜΜΥ, 145Π-58
DESCRIPTION:https://www.ece.tuc.gr/el/katalogos-
 ekdiloseon?tx_tucevents2_tuceventsdi
 splay%5Baction%5D=show&tx_tucevents2
 _tuceventsdisplay%5Bcontroller%5D=Ev
 ent&tx_tucevents2_tuceventsdisplay%5
 Bevent%5D=6705&cHash=34d472039c87f11
 62d6042a50494d17f\nAbstract\n Retrie
 ving videos from the same incident c
 an be formulated as a search-by-exam
 ple problem aiming to discover all v
 ideos in a database related to a giv
 en query. Hence, to tackle this prob
 lem, the following has to be specifi
 ed: (i) what videos are considered r
 elated? (ii) how do we measure simil
 arity between two videos to determin
 e relevance? In this talk, we will d
 elve into our advancements in the vi
 deo retrieval field to provide answe
 rs to both questions. First, we will
  go through our definitions regardin
 g the same incident videos and the c
 omposition of a large-scale dataset 
 of user-generated videos that simula
 te the problem and cover its benchma
 rking needs. Then, we will review ou
 r proposed approaches for the estima
 tion of video similarity. They can b
 e roughly classified into three cate
 gories: (i) Coarse-grained approache
 s that extract global video represen
 tations combined with simple similar
 ity metrics. Our solutions leverage 
 deep Convolutional Neural Networks a
 nd Deep Metric Learning to extract g
 lobal video representations and map 
 videos to feature spaces that preser
 ve video relations. (ii) Fine-graine
 d approaches that employ spatio-temp
 oral representations and similarity 
 functions. Our approach involves a v
 ideo similarity learning network tha
 t captures fine-grained relations be
 tween videos, trained with supervise
 d and self-supervised methodologies.
  (iii) Re-ranking approaches that co
 mbine methods from the two previous 
 categories to perform more efficient
  search. We devise a knowledge disti
 llation scheme that trains two stude
 nt networks based on a teacher, whic
 h are combined with a selector netwo
 rk tuned to achieve an optimal perfo
 rmance-efficiency trade-off. Finally
 , we will benchmark the proposed met
 hods on the composed dataset and com
 pare them with several state-of-the-
 art approaches.\n \n About the Speak
 er\n Dr. Giorgos Kordopatis-Zilos is
  a postdoctoral researcher with the 
 Czech Technical University in Prague
  and part of the Visual Recognition 
 Group (VRG). Prior to that, he was a
  postdoctoral researcher in the Info
 rmation Technologies Institute (ITI)
  of the Centre for Research and Tech
 nology Hellas (CERTH) with the Media
  Analysis, Verification and Retrieva
 l (MeVer) group. In 2013, he receive
 d his diploma in electrical and comp
 uter engineering from the Aristotle 
 University of Thessaloniki (AUTH), G
 reece. In 2021, he was awarded his P
 h.D. from the Queen Mary University 
 of London, UK. His research interest
 s include similarity learning, multi
 media retrieval and matching, superv
 ised and self-supervised learning, k
 nowledge distillation, location esti
 mation, deepfake detection, and forg
 ery localization.\n
STATUS:CONFIRMED
ORGANIZER;RSVP=FALSE;CN=TUC;CUTYPE=TUC:mailto:webmaster@tuc.gr
DTSTART:20240117T110000
DTEND:20240117T121500
TRANSP:OPAQUE
CLASS:DEFAULT
END:VEVENT
END:VCALENDAR