Learning SO(3) Equivariant Representations with Spherical CNNs

Carlos Esteves, Christine Allen-Blanchette, Ameesh Makadia, Kostas Daniilidis

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

19 Scopus citations

Abstract

We address the problem of 3D rotation equivariance in convolutional neural networks. 3D rotations have been a challenging nuisance in 3D classification tasks requiring higher capacity and extended data augmentation in order to tackle it. We model 3D data with multi-valued spherical functions and we propose a novel spherical convolutional network that implements exact convolutions on the sphere by realizing them in the spherical harmonic domain. Resulting filters have local symmetry and are localized by enforcing smooth spectra. We apply a novel pooling on the spectral domain and our operations are independent of the underlying spherical resolution throughout the network. We show that networks with much lower capacity and without requiring data augmentation can exhibit performance comparable to the state of the art in standard retrieval and classification benchmarks.

Original languageEnglish (US)
Title of host publicationComputer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
EditorsVittorio Ferrari, Cristian Sminchisescu, Yair Weiss, Martial Hebert
PublisherSpringer Verlag
Pages54-70
Number of pages17
ISBN (Print)9783030012601
DOIs
StatePublished - 2018
Externally publishedYes
Event15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany
Duration: Sep 8 2018Sep 14 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11217 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other15th European Conference on Computer Vision, ECCV 2018
Country/TerritoryGermany
CityMunich
Period9/8/189/14/18

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Learning SO(3) Equivariant Representations with Spherical CNNs'. Together they form a unique fingerprint.

Cite this