Giorgos Kordopatis-Zilos presents Content-based Video Retrieval: Task Definitions, Datasets and Methods

On 2022-02-10 14:00:00 at G205, Karlovo náměstí 13, Praha 2
Hybrid event: G205 and

Abstract: Content-based video retrieval is a search-by-example retrieval
aiming to retrieve all videos in a database related to a given query. Hence, to
tackle this problem, the following have to be specified: (i) what videos are
considered related? (ii) How to measure similarity between two videos to
determine relevance?
In this talk, we will delve into the literature of the field to provide answers
to both questions. During the first part, we will go through the most prevalent
task definitions that determine the video relations. They adopt various scopes
that range from narrow, where only near-duplicates or videos depicting the same
incident are considered related, to very broad, where videos from the same
or with the same semantics are labelled as relevant. Additionally, based on the
above definitions, several large-scale datasets have been compiled, composed of
user-generated videos. During the second part, we will review the
state-of-the-art methods for the calculation of video similarity. They can be
roughly classified into three categories: (i) coarse-grained approaches that
represent videos as global vectors and use simple functions, e.g., dot-product
or Euclidean distance, to measure similarity, (ii) fine-grained approaches that
employ spatio-temporal representations and similarity calculations, (iii)
re-ranking approaches that combine methods from the two previous categories to
perform retrieval. Finally, we will draw comparisons regarding the methods'
retrieval performance and computational efficiency based on the composed

Bio: Giorgos Kordopatis-Zilos is a Postdoctoral Research Fellow at the
Information Technologies Institute (ITI) of the Centre for Research and
Technology Hellas (CERTH) and a member of the Media Verification (MeVer) team.
He received a Diploma degree in Electrical and Computer Engineering from the
Aristotle University of Thessaloniki (AUTH), Greece, in 2013. Recently, he
defended his Ph.D. dissertation at the Queen Mary University of London on
fine-grained incident video retrieval with video similarity learning. His
research interests include multimedia indexing and retrieval, similarity
learning, knowledge distillation, selective prediction, detection/localization
of manipulated or synthetic media (DeepFakes), image forensics,
visual/text-based location estimation and multimodal learning.
Za obsah zodpovídá: Petr Pošík