Benchmarking

Posted on May 10nd, 2009 by iTowns team

You will find image packages and ground truth for benchmarking as well as the XML format used in results exchanging.
XML example: Face detection results for the image Paris_12-080422_0687-21-00001_0001247
So remember that there is one xml per image (not pano) per features per user. Class can be "Text", "People", "RoadMark", "RoadSign", "Facade" Then each of the class has different types specializing itself.

  • For People: "Face", "Outline"
  • For text: "Line", "Word", "Group"
  • For RoadSign: "Traffic", "Prohibition", "Obligation", "Indication", "Danger"
  • For RoadMark: "LaneSeparator", "CirculationLane", "OtherSigns"
  • For Facade: "MainePlane", "Shop", "Haussman", "Porche"

The polygon description is closed! The attribute is a string that can specify more your feature (color, text description, RoadSign code...)
Confidence is a float between 0 and 1. Verified is 1 for ground truth, 0 for computed.
Access images package for Road Sign

Tags: geographic   image-based   panoramic   content-based   georeferenced   mobile mapping   objet detection   data mining