BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Multi-task learning for autonomous driving
DTSTART:20180820T103000
DTEND:20180820T113000
DTSTAMP:20260429T114708Z
UID:641fa62fa5bf33a2a4a0a061caa2c4eea0ef6d97ff00abc24f2b3491
CATEGORIES:Conferences - Seminars
DESCRIPTION:George Adaimi\nEDIC candidacy exam\nExam president: Prof. Pasc
 al Frossard\nThesis advisor: Prof. Alexandre Massoud Alahi\nCo-examiner: P
 rof. Jean-Philippe Thiran\n\nAbstract\nAutonomous vehicles rely on an accu
 rate perception module. One of the fundamental challenges is to efficientl
 y track pedestrians surrounding a vehicle to anticipate risky situations. 
 Over the past decades\, researchers have formulated the tracking problem a
 s a data association one where they proposed various representations aimin
 g for invariance to nuisances such as viewpoint changes\, body deformation
 \, object occlusion\, and illumination changes. However\, these methods st
 ill suffer to address abrupt changes since they do not explicitly model th
 e nature of the nuisances.\n\nIn this work\, we propose to train a classif
 ier that recognizes these nuisances\, more specifically rotational body de
 formation of pedestrians. We aim to detect deformations as a method to fin
 d a good representation that will lead to better tracking of pedestrians a
 s well as other tasks.\n\nBackground papers\nDynamic Routing between Capsu
 les \, by Sabour\, S.\, et al.\nTaskonomy: Disentangling Task Transfer Lea
 rning\, Zamir\, A.\, et al.\nPerson Reidentification Using Deep Convnets w
 ith Multi-Task Learning\, by McLaughlin\, N.\, et al.\n\n \n 
LOCATION:GC C1 384 https://plan.epfl.ch/?room=GCC1384
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR