Visual Hull Data Sets

Svetlana Lazebnik, Yasutaka Furukawa and Jean Ponce

Beckman Institute and Department of Computer Science
University of Illinois at Urbana-Champaign

Format

Contour

We borrowed the contour data format from Jean-Sebastien Franco's Exact Polyhedral Visual Hull program.

Each contour consists of 2 lines. The first line is just 2 numbers: an image number and the number of points in the contour. The second line is a sequence of image (x, y) coordinates of points along the contour. Note that an image may contain holes, and hence, have multiple contours. Points of a hole are oriented in the opposite direction from those of an outer-silhouette. Also note that not all the holes have been extracted from images and camera parameters may contain errors (not a synthetic data set!). However, the visual hull should be defined given any set of contours and camera parameters, and there should be no problem.

Camera Parameters

A 3x4 projection matrix is given for each image. For a data set of N images, you will find 3N x 4 matrix in the file, which is just an array of 3x4 matrices.

Note: we also provide multiview stereo data sets.

References

Projective Visual Hulls
Svetlana Lazebnik
MS Thesis, University of Illinois at Urbana-Champaign, December 2002.

On Computing Exact Visual Hulls of Solids Bounded by Smooth Surfaces
Svetlana Lazebnik, Edmond Boyer and Jean Ponce
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Kauai, Hawaii, December 2001, Vol. 1, pp. 156-161.

Exact Polyhedral Visual Hulls
Jean-Sebastien Franco, Edmond Boyer
British Machine Vision Conference (BMVC'03)
Vol. I, p. 329-338, September 2003.