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

Kieran Saunders; George Vogiatzis; Luis J. Manso

doi:10.48550/arXiv.2206.03799

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 output › Conference 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 language	English
Title of host publication	2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)
Editors	Ana C. Lopes, Gabriel Pires, Vitor H. Pinto, Jose L. Lima, Pedro Fonseca
Publisher	IEEE
Pages	10-16
Number of pages	7
ISBN (Electronic)	979-8-3503-0121-2
ISBN (Print)	979-8-3503-0122-9
DOIs	https://doi.org/10.48550/arXiv.2206.03799 https://doi.org/10.1109/ICARSC58346.2023.10129564
Publication status	Published - 25 May 2023
Event	2023 IEEE International Conference on Autonomous Robot Systems and Competitions - Tomar, Portugal Duration: 26 Apr 2023 → 27 Apr 2023

Publication series

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

Conference

Conference	2023 IEEE International Conference on Autonomous Robot Systems and Competitions
Abbreviated title	ICARSC 2023
Country/Territory	Portugal
City	Tomar
Period	26/04/23 → 27/04/23

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

Access to Document

Cite this

Saunders, K., Vogiatzis, G., & Manso, L. J. (2023). Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps. In A. C. Lopes, G. Pires, V. H. Pinto, J. L. Lima, & P. Fonseca (Eds.), 2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC) (pp. 10-16). (Proceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)). IEEE. https://doi.org/10.48550/arXiv.2206.03799, https://doi.org/10.1109/ICARSC58346.2023.10129564

Saunders, Kieran ; Vogiatzis, George ; Manso, Luis J. / Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps. 2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC). editor / Ana C. Lopes ; Gabriel Pires ; Vitor H. Pinto ; Jose L. Lima ; Pedro Fonseca. IEEE, 2023. pp. 10-16 (Proceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)).

@inproceedings{033c1bb6af4a42c2b1894b2a9d449fd2,

title = "Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps",

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.",

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",

author = "Kieran Saunders and George Vogiatzis and Manso, {Luis J.}",

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: {\textcopyright} 2023 IEEE.; 2023 IEEE International Conference on Autonomous Robot Systems and Competitions, ICARSC 2023 ; Conference date: 26-04-2023 Through 27-04-2023",

year = "2023",

month = may,

day = "25",

doi = "10.48550/arXiv.2206.03799",

language = "English",

isbn = "979-8-3503-0122-9",

series = "Proceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)",

publisher = "IEEE",

pages = "10--16",

editor = "Lopes, {Ana C.} and Gabriel Pires and Pinto, {Vitor H.} and Lima, {Jose L.} and Pedro Fonseca",

booktitle = "2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)",

address = "United States",

}

Saunders, K, Vogiatzis, G & Manso, LJ 2023, Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps. in AC Lopes, G Pires, VH Pinto, JL Lima & P Fonseca (eds), 2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC). Proceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC), IEEE, pp. 10-16, 2023 IEEE International Conference on Autonomous Robot Systems and Competitions, Tomar, Portugal, 26/04/23. https://doi.org/10.48550/arXiv.2206.03799, https://doi.org/10.1109/ICARSC58346.2023.10129564

Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps. / Saunders, Kieran; Vogiatzis, George; Manso, Luis J.
2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC). ed. / Ana C. Lopes; Gabriel Pires; Vitor H. Pinto; Jose L. Lima; Pedro Fonseca. IEEE, 2023. p. 10-16 (Proceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)).

Research output: Chapter in Book/Published conference output › Conference publication

TY - GEN

T1 - Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps

AU - Saunders, Kieran

AU - Vogiatzis, George

AU - Manso, Luis J.

PY - 2023/5/25

Y1 - 2023/5/25

N2 - 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.

AB - 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.

KW - Training, Computer vision, Image analysis, Estimation, Graphics processing units, Computer architecture, Complexity theory

KW - Computer vision, Autonomous vehicles, 3D/stereo scene analysis, Vision and Scene Understanding

UR - https://ieeexplore.ieee.org/document/10129564

U2 - 10.48550/arXiv.2206.03799

DO - 10.48550/arXiv.2206.03799

M3 - Conference publication

SN - 979-8-3503-0122-9

T3 - Proceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)

SP - 10

EP - 16

BT - 2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)

A2 - Lopes, Ana C.

A2 - Pires, Gabriel

A2 - Pinto, Vitor H.

A2 - Lima, Jose L.

A2 - Fonseca, Pedro

PB - IEEE

T2 - 2023 IEEE International Conference on Autonomous Robot Systems and Competitions

Y2 - 26 April 2023 through 27 April 2023

ER -

Saunders K, Vogiatzis G, Manso LJ. Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps. In Lopes AC, Pires G, Pinto VH, Lima JL, Fonseca P, editors, 2023 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC). IEEE. 2023. p. 10-16. (Proceedings of IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)). doi: 10.48550/arXiv.2206.03799, 10.1109/ICARSC58346.2023.10129564

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

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this