PerformingNon-Rigid Structure fromMotion - Deformable Surface 3D Reconstruction from Monocular Images - page 68

50

0 0 0 0 0 0

0

−50 −50 −50 −50 −50 −50 −50

−100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100

x x x x x

x x

100 100 100 100 100 100 100

50 50 50 50 50 50 50

0 0 0 0 0 0 0

−50 −50 −50 −50 −50 −50 −50

−100 −100 −100 −100 −100 −100 −100

−100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100

x x x x x x x

100 100 100 100 100 100 100

50 50 50 50 50 50 50

0 0 0 0 0 0 0

−50 −50 −50 −50 −50 −50 −50

−100 −100 −100 −100 −100 −100 −100

−100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100

x x x x x x x

100 100 100 100 100 100

100

50 50 50 50 50 50

50

0 0 0 0 0 0

0

−50 −50 −50 −50 −50 −50

−50

−100 −100 −100 −100 −100 −100 −100

−100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100

x x x x x x x

100 100 100 100 100 100 100

50 50 50 50 50 50 50

0 0 0 0 0 0 0

−50 −50 −50 −50 −50 −50 −50

−100 −100 −100 −100 −100 −100 −100

−100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100

x x x x x x x

100 100 100 100 100 100 100

50 50 50 50 50 50 50

0 0 0 0 0 0 0

−50 −50 −50 −50 −50 −50 −50

−100 −100 −100 −100 −100 −100 −100

−100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100

x x x x x x x

t=20 t=50 t=80 t=115 t=148 t=175 t=200

Figure 6.7: Comparison of the results of the various algorithms proposed in Torresani et al. [ 2008 ]
(BCD-LS, EM-PPCA, EM-LDS), as well as of the methods in Xiao et al. [ 2004b ] (XCK) and in Brand
[ 2005 ] (B05) on the shark data. Green circles are the ground-truth points, and blue dots the reconstructed
ones. Courtesy of A. Hertzmann. © 2008 IEEE.

introduced. It starts by computing the mean shape, and then iteratively adds modes to explain as
much of the remaining variance of the data as possible. A stopping criterion based on cross-validation
was defined to avoid overfitting to measurement noise. In Brandt et al. [ 2009 ], it was proposed to
find independent basis shapes by following an ICA-based approach. To this end, a regularization
term that minimizes the mutual information of the individual modes was introduced.