BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:Europe/Paris
X-WR-TIMEZONE:Europe/Paris
BEGIN:VEVENT
UID:6395@i2m.univ-amu.fr
DTSTART;TZID=Europe/Paris:20210604T143000
DTEND;TZID=Europe/Paris:20210604T153000
DTSTAMP:20241120T201420Z
URL:https://www.i2m.univ-amu.fr/evenements/robust-image-representation-for
 -classification-and-object-discovery/
SUMMARY:Oriane Siméoni (INRIA Rennes-Bretagne Atlantique): Robust image re
 presentation for classification and object discovery
DESCRIPTION:Oriane Siméoni: Neural network representations proved to be re
 levant for many computer vision tasks such as image classification\, objec
 t detection\, segmentation or instance-level image retrieval. A network is
  trained for one particular task and requires a large number of labeled da
 ta. We propose in this thesis solutions to extract the most information wi
 th the least supervision. First focusing on the classification task\, we e
 xamine the active learning process in the context of deep learning and sho
 w that combining it to semi-supervised and unsupervised techniques boost g
 reatly results. We then investigate the image retrieval task\, and in part
 icular we exploit the spatial localization information available "for free
 '' in CNN feature maps. We discover objects of interest in images of a dat
 aset and gather their representations in a nearest neighbor graph. Using t
 he centrality measure on the graph\, we are able to construct a saliency m
 ap per image which focuses on the repeating objects and allows us to compu
 te a global representation excluding clutter and background. \nhttps://tel
 .archives-ouvertes.fr/tel-03082952\n\n&nbsp\;
ATTACH;FMTTYPE=image/jpeg:https://www.i2m.univ-amu.fr/wp-content/uploads/2
 021/03/Oriane_Simeoni.jpg
CATEGORIES:Séminaire,Signal et Apprentissage
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/Paris
X-LIC-LOCATION:Europe/Paris
BEGIN:DAYLIGHT
DTSTART:20210328T030000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
END:VCALENDAR