Shape Completion Using Deep Boltzmann Machine

Algorithm 2

Find the proper mask for sampling.

Given: A training set consists of samples denoted as . The parameters of the trained DBM .

A new sample with incomplete shape.

Output: A proper mask corresponding to the missing region of

() Train the regression model (9) to obtain the parameters .

() Use (3) for a few of iterations to get an intermediate sample

() Use (7), (8) combined with to obtain.

() Obtain by (10).

() Set the final mask via (11) and (12).