BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//jEvents 2.0 for Joomla//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
UID:rifi9kd23ckudgvor04butk44s@google.com
CATEGORIES:RTportal.ru
CREATED:20170606T092329
SUMMARY:семинар CV: Seed, Expand and Constrain: Three Principles for Weakly-Supervised Image Segmentation
LOCATION:ШАД в Яндексе, аудитория Гарвард
DESCRIPTION;ENCODING=QUOTED-PRINTABLE: Александр Колесников из IST(<a href="http://pub.ist.ac.at/~akolesnikov/">h
 ttp://pub.ist.ac.at/~akolesnikov/</a>)<br/><br/>Title: "Seed, Expand and Co
 nstrain: Three Principles for Weakly-Supervised<br/>Image Segmentation"<br/
 ><br/>Abstract: "We introduce a new loss function for the weakly-supervised
 <br/>training of semantic image segmentation models based on three guiding<
 br/>principles: to seed with weak location cues, to expand objects based on
 <br/>the information about which classes can occur, and to constrain the<br
 />segmentations to coincide with image boundaries. We show experimentally<b
 r/>that training a deep convolutional neural network using the proposed los
 s<br/>function leads to substantially better segmentations than previous<br
 />state-of-the-art methods on the challenging PASCAL VOC 2012 dataset. We<b
 r/>furthermore give insight into the working mechanism of our method by a<b
 r/>detailed experimental study that illustrates how the segmentation qualit
 y<br/>is affected by each term of the proposed loss function as well as the
 ir<br/>combinations."<br/><br/>Запись видео скорее всего будет.<br/>Семинар
 ы проходят в помещениях ШАД в Яндексе, аудитория Гарвард, начало в 16:30. П
 риходить можно всем желающим, но если у вас нет пропуска, то нужно заранее 
 написать мне в fb и подойти к торцу здания Тимура Фрунзе 11с5 за 5 минут до
  начала семинара. Кто планирует ходить постоянно напишите Артему, чтобы зап
 устить процесс создания пропусков (см емейл  This event was imported from: 
 https://rtportal.ru/index.php/kalendar-sobytij/eventdetail/131/-/seminar-cv
 -seed-expand-and-constrain-three-principles-for-weakly-supervised-image-seg
 mentation?tmpl=component
DTSTAMP:20260812T201355Z
DTSTART;TZID=Europe/Moscow:20160323T163000
DTEND;TZID=Europe/Moscow:20160323T180000
SEQUENCE:0
TRANSP:OPAQUE
END:VEVENT
END:VCALENDAR