Competition vs. concatenation in skip connections of fully convolutional networks

Estrada S, Conjeti S, Ahmad M, Navab N, Reuter M (2018)

Publication Type: Conference contribution

Publication year: 2018

Journal

Lecture Notes in Computer Science Springer Verlag

Publisher: Springer Verlag

Book Volume: 11046 LNCS

Pages Range: 214-222

Conference Proceedings Title: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Event location: Granada, ESP

ISBN: 9783030009182

DOI: 10.1007/978-3-030-00919-9_25

Abstract

Increased information sharing through short and long-range skip connections between layers in fully convolutional networks have demonstrated significant improvement in performance for semantic segmentation. In this paper, we propose Competitive Dense Fully Convolutional Networks (CDFNet) by introducing competitive maxout activations in place of naïve feature concatenation for inducing competition amongst layers. Within CDFNet, we propose two architectural contributions, namely competitive dense block (CDB) and competitive unpooling block (CUB) to induce competition at local and global scales for short and long-range skip connections respectively. This extension is demonstrated to boost learning of specialized sub-networks targeted at segmenting specific anatomies, which in turn eases the training of complex tasks. We present the proof-of-concept on the challenging task of whole body segmentation in the publicly available VISCERAL benchmark and demonstrate improved performance over multiple learning and registration based state-of-the-art methods.

Involved external institutions

Deutsches Zentrum für Neurodegenerative Erkrankungen (DZNE) DE Germany (DE) Technische Universität München (TUM) DE Germany (DE)

How to cite

APA:

Estrada, S., Conjeti, S., Ahmad, M., Navab, N., & Reuter, M. (2018). Competition vs. concatenation in skip connections of fully convolutional networks. In Mingxia Liu, Heung-Il Suk, Yinghuan Shi (Eds.), Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (pp. 214-222). Granada, ESP: Springer Verlag.

MLA:

Estrada, Santiago, et al. "Competition vs. concatenation in skip connections of fully convolutional networks." Proceedings of the 9th International Workshop on Machine Learning in Medical Imaging, MLMI 2018 held in conjunction with the 21st International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2018, Granada, ESP Ed. Mingxia Liu, Heung-Il Suk, Yinghuan Shi, Springer Verlag, 2018. 214-222.

BibTeX: Download