Lipschitz Constrained Neural Networks for Robust Object Detection at Sea

Jonathan Binner Becktor; Frederik Emil Thorsson Schöller; Evangelos Boukas; Mogens Blanke; Lazaros Nalpantidis

doi:10.1088/1757-899X/929/1/012023

Lipschitz Constrained Neural Networks for Robust Object Detection at Sea

Jonathan Binner Becktor^*, Frederik Emil Thorsson Schöller, Evangelos Boukas, Mogens Blanke, Lazaros Nalpantidis

^*Corresponding author for this work

Research output: Contribution to journal › Conference article › Research › peer-review

23 Downloads (Pure)

Abstract

Autonomous ships relies on sensory data to perceive other objects of interest in their environment. Deep Learning based object detection in the image domain is a common approach to solve this issue. The robustness of such approaches in non-ideal conditions is, however, still to be proven. In this work state of the art methods are applied on the RetinaNet architecture attempting to create a more robust object detection network given noisy input data. The GroupSort activation function and Spectral Normalization is used and the results are compared to the standard RetinaNet network. Our findings show that these modifications perform better and ensure robustness under moderate noise levels, than the standard RetinaNet network.

Original language	English
Article number	012023
Journal	IOP Conference Series: Materials Science and Engineering
Volume	929
Number of pages	11
ISSN	1757-8981
DOIs	https://doi.org/10.1088/1757-899X/929/1/012023
Publication status	Published - 2020
Event	3^rd International Conference on Maritime Autonomous Surface Ship - Virtual event, Ulsan, Korea, Republic of Duration: 11 Nov 2020 → 12 Nov 2020 Conference number: ICMASS 2020 https://www.icmass-conf.org/

Conference

Conference	3^rd International Conference on Maritime Autonomous Surface Ship
Number	ICMASS 2020
Location	Virtual event
Country	Korea, Republic of
City	Ulsan
Period	11/11/2020 → 12/11/2020
Internet address	https://www.icmass-conf.org/

Access to Document

10.1088/1757-899X/929/1/012023

FulltextFinal published version, 3.25 MB

Fingerprint Dive into the research topics of 'Lipschitz Constrained Neural Networks for Robust Object Detection at Sea'. Together they form a unique fingerprint.

1 Active

ShippingLab Autonomy
Blanke, M., Galeazzi, R., Dittmann, K., Hansen, S., Papageorgiou, D., Nalpantidis, L., Schöller, F. E. T., Plenge-Feidenhans'l, M. K., Hansen, P. N., Andersen, R. H., Becktor, J. B., Enevoldsen, T. T., Dagdilelis, D., Karstensen, P. I. H., Nielsen, R. E., Garde, J. & Ravn, O.
01/04/2019 → 31/12/2022
Project: Research

Cite this

@inproceedings{5274d0180a1940be80f234ce3e934282,

title = "Lipschitz Constrained Neural Networks for Robust Object Detection at Sea",

abstract = "Autonomous ships relies on sensory data to perceive other objects of interest in their environment. Deep Learning based object detection in the image domain is a common approach to solve this issue. The robustness of such approaches in non-ideal conditions is, however, still to be proven. In this work state of the art methods are applied on the RetinaNet architecture attempting to create a more robust object detection network given noisy input data. The GroupSort activation function and Spectral Normalization is used and the results are compared to the standard RetinaNet network. Our findings show that these modifications perform better and ensure robustness under moderate noise levels, than the standard RetinaNet network.",

author = "Becktor, {Jonathan Binner} and Sch{\"o}ller, {Frederik Emil Thorsson} and Evangelos Boukas and Mogens Blanke and Lazaros Nalpantidis",

year = "2020",

doi = "10.1088/1757-899X/929/1/012023",

language = "English",

volume = "929",

journal = "I O P Conference Series: Materials Science and Engineering",

issn = "1757-8981",

publisher = "IOP Publishing",

note = "3<sup>rd</sup> International Conference on Maritime Autonomous Surface Ship ; Conference date: 11-11-2020 Through 12-11-2020",

url = "https://www.icmass-conf.org/",

}

Lipschitz Constrained Neural Networks for Robust Object Detection at Sea. / Becktor, Jonathan Binner ; Schöller, Frederik Emil Thorsson ; Boukas, Evangelos ; Blanke, Mogens ; Nalpantidis, Lazaros.

In: IOP Conference Series: Materials Science and Engineering, Vol. 929, 012023, 2020.

Research output: Contribution to journal › Conference article › Research › peer-review

TY - GEN

T1 - Lipschitz Constrained Neural Networks for Robust Object Detection at Sea

AU - Becktor, Jonathan Binner

AU - Schöller, Frederik Emil Thorsson

AU - Boukas, Evangelos

AU - Blanke, Mogens

AU - Nalpantidis, Lazaros

N1 - Conference code: ICMASS 2020

PY - 2020

Y1 - 2020

N2 - Autonomous ships relies on sensory data to perceive other objects of interest in their environment. Deep Learning based object detection in the image domain is a common approach to solve this issue. The robustness of such approaches in non-ideal conditions is, however, still to be proven. In this work state of the art methods are applied on the RetinaNet architecture attempting to create a more robust object detection network given noisy input data. The GroupSort activation function and Spectral Normalization is used and the results are compared to the standard RetinaNet network. Our findings show that these modifications perform better and ensure robustness under moderate noise levels, than the standard RetinaNet network.

AB - Autonomous ships relies on sensory data to perceive other objects of interest in their environment. Deep Learning based object detection in the image domain is a common approach to solve this issue. The robustness of such approaches in non-ideal conditions is, however, still to be proven. In this work state of the art methods are applied on the RetinaNet architecture attempting to create a more robust object detection network given noisy input data. The GroupSort activation function and Spectral Normalization is used and the results are compared to the standard RetinaNet network. Our findings show that these modifications perform better and ensure robustness under moderate noise levels, than the standard RetinaNet network.

U2 - 10.1088/1757-899X/929/1/012023

DO - 10.1088/1757-899X/929/1/012023

M3 - Conference article

VL - 929

JO - I O P Conference Series: Materials Science and Engineering

JF - I O P Conference Series: Materials Science and Engineering

SN - 1757-8981

M1 - 012023

T2 - 3<sup>rd</sup> International Conference on Maritime Autonomous Surface Ship

Y2 - 11 November 2020 through 12 November 2020

ER -

Lipschitz Constrained Neural Networks for Robust Object Detection at Sea

Abstract

Conference

Access to Document

ShippingLab Autonomy

Cite this