Deep Part-Based Generative Shape Model with Latent Variables

Alexander Kirillov (1) , Mikhail Gavrikov (2) , Ekaterina Lobacheva (3) , Anton Osokin (4, 5) , Dmitry Vetrov (3)

1 TU Dresden - Technische Universität Dresden = Dresden University of Technology
2 MSU - Lomonosov Moscow State University
3 CS-HSE - Faculty of Computer Science [Moscow]
4 SIERRA - Statistical Machine Learning and Parsimony
5 MSR - INRIA - Microsoft Research - Inria Joint Centre

The Shape Boltzmann Machine (SBM) [6] and its multilabel version MSBM [5] have been recently introduced as deep generative models that capture the variations of an object shape. While being more flexible MSBM requires datasets with labeled parts of the objects for training. In the paper we present an algorithm for training MSBM using binary masks of objects and the seeds which approximately correspond to the locations of objects parts. The latter can be obtained from part-based detectors in an unsupervised manner. We derive a latent variable model and an EM-like training procedure for adjusting the weights of MSBM using a deep learning framework. We show that the model trained by our method outperforms SBM in the tasks related to binary shapes and is very close to the original MSBM in terms of quality of multilabel shapes.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Alexander Kirillov, Mikhail Gavrikov, Ekaterina Lobacheva, Anton Osokin, Dmitry Vetrov. Deep Part-Based Generative Shape Model with Latent Variables. 27th British Machine Vision Conference (BMVC 2016), Sep 2016, York, United Kingdom. ⟨hal-01404071⟩