Shape Completion Using Deep Boltzmann Machine

Algorithm 3

Shape completion.

Given: A training set consists of samples denoted as . A new sample with incomplete shape.

Output: A new sample with complete shape.

() Use Algorithm 1 to train a DBM.

() Use Algorithm 2 to set a proper mask .

() Randomly clamp the state of hidden layer , clamp the visible state using mask .

() With the clamped units, sample from the DBM using the indicator .