@inproceedings{Gronat-CVPR-2013,
  IS = { zkontrolovano 20 Jan 2014 },
  UPDATE  = { 2014-01-20 },
  author = {Gron{\' a}t, Petr and Obozinski, Guillume and {\v S}ivic, Josef and Pajdla, Tom{\' a}{\v
 s}},
  title = {Learning and Calibrating Per-Location Classifiers for Visual Place Recognition},
  booktitle = {CVPR: 2013 IEEE Computer Society Conference on Computer 
    Vision and Pattern Recognition},
  book_pages = {3735},
  pages = {907--914},
  publisher = {IEEE Computer Society},
  address = {Los Alamitos, USA},
  isbn = {978-0-7695-4989-7},
  issn = {1063-6919},
  year = {2013},
  month = {June},
  day = {25-27},
  venue = {Portland, US},
  annote = {The aim of this work is to localize a query photograph by finding other images dep
icting the same place in a large geotagged image database. This is a challenging task due to c
hanges in viewpoint, imaging conditions and the large size of the image database. The contribu
tion of this work is two-fold. First, we cast the place recognition problem as a classificatio
n task and use the available geotags to train a classifier for each location in the database i
n a similar manner to per-exemplar SVMs in object recognition. Second, as only few positive tr
aining examples are available for each location, we propose a new approach to calibrate all th
e per-location SVM classifiers using only the negative examples. The calibration we propose re
lies on a significance measure essentially equivalent to the p-values classically used in stat
istical hypothesis testing. Experiments are performed on a database of 25,000 geotagged street
 view images of Pittsburgh and demonstrate improved place recognition accuracy of the proposed
 approach over the previous work.},
  keywords = {image based localization, SVM classifier, classifier calibration},
  prestige = {important},
  note = {CD-ROM},
  psurl = {[Gronat-CVPR-2013.pdf]},
authorship =  {25-25-25-25},
  project = {FP7-SPACE-312377 PRoViDE,SGS13/140/OHK3/2T/13},
}