Introduction


When recognizing objects with a mobile platform, visual appearance is only one of the many different sources of information available. Robot proprioception, 3D sensing of various kinds, and higher-level spatial context such as what room in the building are often also available to a robot. The scientific community is recently embracing the use of many non-appearance cues to help with recognition, and this dataset attempts to capture the unique set of data that is available to a mobile platform, for a sampling of realistic and challenging scenes, with a large enough data volume and quality to facilitate careful study of this problem


This page contains a description of the data that has been collected, as well as download information that will allow other authors to obtain a copy for their own study. Thank you for visiting this page, and please do take a look at the data if it is of interest.

VRS in ACCV 2010:

An image-only branch of the dataset was used for evaluation of multi-view scene labeling in a paper at ACCV 2010 in New Zealand.

Paper

Initial data released!

Version 1.0: September 29th, 2010. This version of the data contains images used for ACCV as well as scenes collected by a robot that include sensed point clouds.

Download

Code Release

All source code related to this dataset, including the code used to move the robot platform and collect sensor readings, the registration code, the annotation GUI and our latest inference methods is online:

ubc-ros-pkg