Towards deep compositional networks

Domen Tabernik, Matej Kristan, Jeremy Wyatt, Ales Leonardis

Research output: Chapter in Book/Report/Conference proceedingConference contribution

8 Citations (Scopus)
261 Downloads (Pure)


Hierarchical feature learning based on convolutional neural networks (CNN) has recently shown significant potential in various computer vision tasks. While allowing high-quality discriminative feature learning, the downside of CNNs is the lack of explicit structure in features, which often leads to overfitting, absence of reconstruction from partial observations and limited generative abilities. Explicit structure is inherent in hierarchical compositional models, however, these lack the ability to optimize a well-defined cost function. We propose a novel analytic model of a basic unit in a layered hierarchical model with both explicit compositional structure and a well-defined discriminative
cost function. Our experiments on two datasets show that the proposed compositional model performs on a par with standard CNNs on discriminative tasks, while, due to explicit modeling of the structure in the feature units, affording a straight-forward visualization of parts and faster inference due to separability of
the units.
Original languageEnglish
Title of host publicationProceedings ICPR 2016
Subtitle of host publication23rd International Conference on Pattern Recognition
PublisherIEEE Computer Society
Number of pages6
ISBN (Electronic)978-1-5090-4847-2
ISBN (Print)978-1-5090-4848-9
Publication statusPublished - 24 Apr 2017
Event23rd International Conference on Pattern Recognition - Cancun, Mexico
Duration: 4 Dec 20168 Dec 2016


Conference23rd International Conference on Pattern Recognition


Dive into the research topics of 'Towards deep compositional networks'. Together they form a unique fingerprint.

Cite this