BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Rich semantic representations for detailed visual recognition - Su
 bhransu Maji\, Toyota Technological Institute at Chicago
DTSTART:20140326T110000Z
DTEND:20140326T120000Z
UID:TALK51617@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:Several problems in computer vision can be cast as a mapping f
 rom input (e.g.\, images and video) to richly structured spaces (e.g.\, at
 tributes\, 3D layout\, and pose). Often the choice of the underlying repre
 sentation of the input is crucial to the success of automatic methods for 
 such mappings. On one hand\, representations that are semantically aligned
  can enable better human-centric applications\, but on the other hand\, re
 presentations that are not necessarily semantic when learned from `big-dat
 a’ tends to have better empirical performance.\n\nI’ll show that with 
 a careful design of the learning/inference method and small amounts of add
 itional supervision\, one can learn representations that achieve both the 
 goals. Our methods leverage noisy annotations collected via “crowdsourci
 ng” to discover semantically aligned representations that enable several
  high-level recognition tasks. In particular\, we achieve state of the art
  results for person detection and attribute recognition on the PASCAL VOC 
 datasets\, and material recognition on the KTH-TIPS/Flickr datasets. I’l
 l also present instances where algorithms consider humans “in the loop
 ” to solve challenging tasks\, such as\, fine-grained category recogniti
 on (e.g. is this bird a Quetzal?)\, discriminative part/attribute discover
 y\, and to enable faster annotation interfaces.\n
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station Road\, Cambridge
 \, CB1 2FB
END:VEVENT
END:VCALENDAR