Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps

Kieran Saunders*, George Vogiatzis, Luis J. Manso

*Corresponding author for this work

Research output: Chapter in Book/Published conference outputConference publication

Abstract

Self-supervised monocular depth estimation has been a subject of intense study in recent years, because of its applications in robotics and autonomous driving. Much of the recent work focuses on improving depth estimation by increasing architecture complexity. This paper shows that state-of-the-art performance can also be achieved by improving the learning process rather than increasing model complexity. More specifically, we propose (i) disregarding small potentially dynamic objects when training, and (ii) employing an appearance-based approach to separately estimate object pose for truly dynamic objects. We demonstrate that these simplifications reduce GPU memory usage by 29% and result in qualitatively and quantitatively improved depth maps.
Original languageEnglish
Title of host publication2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)
EditorsAna C. Lopes, Gabriel Pires, Vitor H. Pinto, Jose L. Lima, Pedro Fonseca
PublisherIEEE
Pages10-16
Number of pages7
ISBN (Electronic)979-8-3503-0121-2
ISBN (Print)979-8-3503-0122-9
DOIs
Publication statusPublished - 25 May 2023
Event2023 IEEE International Conference on Autonomous Robot Systems and Competitions - Tomar, Portugal
Duration: 26 Apr 202327 Apr 2023

Publication series

NameProceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)
PublisherIEEE
ISSN (Print)2573-9360
ISSN (Electronic)2573-9387

Conference

Conference2023 IEEE International Conference on Autonomous Robot Systems and Competitions
Abbreviated titleICARSC 2023
Country/TerritoryPortugal
CityTomar
Period26/04/2327/04/23

Bibliographical note

Funding Information:
Most experiments were run on Aston EPS Machine Learning Server, funded by the EPSRC Core Equipment Fund, Grant EP/V036106/1.

Publisher Copyright:
© 2023 IEEE.

Keywords

  • Training, Computer vision, Image analysis, Estimation, Graphics processing units, Computer architecture, Complexity theory
  • Computer vision, Autonomous vehicles, 3D/stereo scene analysis, Vision and Scene Understanding

Fingerprint

Dive into the research topics of 'Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps'. Together they form a unique fingerprint.

Cite this