BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Microsoft Research Cambridge\, public talks
SUMMARY:Large-Scale Camera Pose Voting and the Geometric B
 urstiness Problem - Torsten Sattler\, ETH Zurich
DTSTART;TZID=Europe/London:20160419T110000
DTEND;TZID=Europe/London:20160419T120000
UID:TALK65605AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/65605
DESCRIPTION:Location recognition is the problem of determining
  the place depicted in a given photo. In the first
  part of the talk\, we consider the case where the
  scene is represented by a 3D model and we are not
  only interested in the place depicted in an image
  but also the position and orientation from which 
 the image was taken\, i.e.\, the camera pose. A ma
 jor challenge for solving this image-based localiz
 ation problem is to establish the 2D-3D matches re
 quired for pose estimation. This is especially tru
 e for large scale scenes containing many 3D points
  with similar local appearance. Instead of using e
 laborate matching schemes\, we introduce an effici
 ent camera voting approach whose run-time is indep
 endent of the inlier ratio. While our approach all
 ows us to handle an arbitrary number of matches in
  linear time\, we show that simply increasing the 
 number of 2D-3D matches used for pose estimation d
 oes not solve the image-based localization problem
 . The reason for this behavior is that increasing 
 the number of matches leads to wrong poses with ma
 ny inliers. In the second part of the talk\, we th
 us consider the problem of finding a better decisi
 on criterion than the raw inlier count. We show th
 at geometric bursts\, i.e.\, spatial configuration
 s appearing at multiple places in a scene\, are a 
 major reason why the raw inlier count fails for la
 rge-scale location recognition. We introduce simpl
 e schemes that allow us to efficiently detect geom
 etric bursts during query time. We show experiment
 ally that down-weighting inliers based on the numb
 er of bursts they appear in allows us to better de
 cide between correct and incorrect place recogniti
 on results and significantly boosts the location r
 ecognition performance.\n\n
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station R
 oad\, Cambridge\, CB1 2FB
CONTACT:
END:VEVENT
END:VCALENDAR