Datasets for ECCV 2004 (Brand, Antone, Teller)

Overview

This directory contains three datasets (each containing three files) to accompany our ECCV 2004 paper (Matthew Brand, Matthew Antone, and Seth Teller, "Spectral solution of large-scale extrinsic camera calibration as a graph embedding problem," ECCV 2004, to appear).

Datasets in the "rot" directories represent nodes with positions derived from raw GPS, and orientations (not given here, but available from the City Scanning Project Dataset) derived from VP (vanishing point) alignment.

Datasets in the "trans" directories represent nodes with positions derived from Antone and Teller's linear optimization scheme (see IJCV 2002 and 2003 papers here) using inter-node baselines, and identical orientations to the "rot" nodes.

Within each dataset, there are three files:

pos*.txt, which has one line per node, containing the integer node ID, the token "Pos", and a three-vector (x,y,z) for the node position;

constraints*.m (in matlab format), which holds two arrays from_node[] and to_node[] of integer (node) indices, and an array direction[] of unitized 3-vectors; and

pos*Revised.txt, with revised position estimates, in the same format as the file of unrevised positions.

N.B.: some node ID's are absent, so the file line number is not in general equivalent to the node ID!

Green Building Dataset (32 nodes, 80 edges)

Positions: green_rot/posGreen.txt

Constraints: green_rot/constraintsGreen.m

Revised positions: green_trans/posGreenRevised.txt

Ames Court Dataset (158 nodes, 355 edges)

Positions: ames_rot/posAmes.txt

Constraints: ames_rot/constraintsAmes.m

Revised positions: ames_trans/posAmesRevised.txt

All Campus Dataset (566 nodes, 1122 edges)

Positions: campus_rot/posCampus.txt

Constraints: campus_rot/constraintsCampus.m

Revised positions: campus_trans/posCampusRevised.txt

Please email us if you have any problems making sense of this data.