A Quasi-Dense Approach to Surface Reconstruction from Uncalibrated Images
Maxime Lhuillier and Long Quan.
IEEE Transactions on Pattern Analysis and Machine Intelligence,
 27(3):418-433, 2005.

DATA DESCRIPTION
This page contains data used or produced by the method described in the
 paper above, including
- the image sequence
- reconstructed cameras (a list of 3x4 matrix)
- reconstructed points (a list of 3D points)
- the silhouettes if needed
- the 3D model (a VRML 1.0 ASCII file and a JPEG file for the texture)

MORE DETAILS
The cameras and points are the results of the Euclidian bundle adjustment.
- Camera are the usual 3x4 matrices defined up to a scale, such that
 pixel (0,0) is the center of the pixel at the top left corner.
The file begin with the line "NbCameras x", with x the number of cameras.
- Points are those of the complete scene, including foreground and background
 (they are not filtered).
The file begin with the line "NbPoints x", with x the number of points.
- Silhouettes files are ASCII files which contain a list of region, one for
 each image of the sequence.
For each region, the first line contains the x and y size of the image, the
 last line is -1, and intermediate lines have the form "y n x1 X1 ..... xn Xn"
with y the number of the slicing row of the image (0 is the top row), n is the
 number of interval [x1, X1[, [x2, X2[, .... [xn Xn[ such that the intersection
 of the region and the y-th row are pixels of abscissa in [xi Xi[.

COPYRIGHTS (C) 2005
You are free to download and use these data for research purposes only,
 and acknowledge the reference.