Standard

How Do Neural Networks See Depth in Single Images? / van Dijk, Tom; de Croon, Guido.

The IEEE International Conference on Computer Vision (ICCV). 2019. p. 2183-2191.

Research output: Chapter in Book/Conference proceedings/Edited volumeConference contributionScientificpeer-review

Harvard

van Dijk, T & de Croon, G 2019, How Do Neural Networks See Depth in Single Images? in The IEEE International Conference on Computer Vision (ICCV). pp. 2183-2191, The IEEE International Conference on Computer Vision 2019, Seoul, Korea, Republic of, 27/10/19. <http://openaccess.thecvf.com/content_ICCV_2019/html/van_Dijk_How_Do_Neural_Networks_See_Depth_in_Single_Images_ICCV_2019_paper.html>

APA

Vancouver

van Dijk T, de Croon G. How Do Neural Networks See Depth in Single Images? In The IEEE International Conference on Computer Vision (ICCV). 2019. p. 2183-2191

Author

van Dijk, Tom ; de Croon, Guido. / How Do Neural Networks See Depth in Single Images?. The IEEE International Conference on Computer Vision (ICCV). 2019. pp. 2183-2191

BibTeX

@inproceedings{fbe6b0d4cd10450197da2e6f77f7f4cd,
title = "How Do Neural Networks See Depth in Single Images?",
abstract = "Deep neural networks have lead to a breakthrough in depth estimation from single images. Recent work shows that the quality of these estimations is rapidly increasing. It is clear that neural networks can see depth in single images. However, to the best of our knowledge, no work currently exists that analyzes what these networks have learned. In this work we take four previously published networks and investigate what depth cues they exploit. We find that all networks ignore the apparent size of known obstacles in favor of their vertical position in the image. The use of the vertical position requires the camera pose to be known; however, we find that these networks only partially recognize changes in camera pitch and roll angles. Small changes in camera pitch are shown to disturb the estimated distance towards obstacles. The use of the vertical image position allows the networks to estimate depth towards arbitrary obstacles - even those not appearing in the training set - but may depend on features that are not universally present.",
keywords = "neural networks, monocular depth estimation, Depth perception",
author = "{van Dijk}, Tom and {de Croon}, Guido",
year = "2019",
month = oct,
language = "English",
pages = "2183--2191",
booktitle = "The IEEE International Conference on Computer Vision (ICCV)",
note = "The IEEE International Conference on Computer Vision 2019, ICCV ; Conference date: 27-10-2019 Through 02-11-2019",
url = "http://iccv2019.thecvf.com/",

}

RIS

TY - GEN

T1 - How Do Neural Networks See Depth in Single Images?

AU - van Dijk, Tom

AU - de Croon, Guido

PY - 2019/10

Y1 - 2019/10

N2 - Deep neural networks have lead to a breakthrough in depth estimation from single images. Recent work shows that the quality of these estimations is rapidly increasing. It is clear that neural networks can see depth in single images. However, to the best of our knowledge, no work currently exists that analyzes what these networks have learned. In this work we take four previously published networks and investigate what depth cues they exploit. We find that all networks ignore the apparent size of known obstacles in favor of their vertical position in the image. The use of the vertical position requires the camera pose to be known; however, we find that these networks only partially recognize changes in camera pitch and roll angles. Small changes in camera pitch are shown to disturb the estimated distance towards obstacles. The use of the vertical image position allows the networks to estimate depth towards arbitrary obstacles - even those not appearing in the training set - but may depend on features that are not universally present.

AB - Deep neural networks have lead to a breakthrough in depth estimation from single images. Recent work shows that the quality of these estimations is rapidly increasing. It is clear that neural networks can see depth in single images. However, to the best of our knowledge, no work currently exists that analyzes what these networks have learned. In this work we take four previously published networks and investigate what depth cues they exploit. We find that all networks ignore the apparent size of known obstacles in favor of their vertical position in the image. The use of the vertical position requires the camera pose to be known; however, we find that these networks only partially recognize changes in camera pitch and roll angles. Small changes in camera pitch are shown to disturb the estimated distance towards obstacles. The use of the vertical image position allows the networks to estimate depth towards arbitrary obstacles - even those not appearing in the training set - but may depend on features that are not universally present.

KW - neural networks

KW - monocular depth estimation

KW - Depth perception

M3 - Conference contribution

SP - 2183

EP - 2191

BT - The IEEE International Conference on Computer Vision (ICCV)

T2 - The IEEE International Conference on Computer Vision 2019

Y2 - 27 October 2019 through 2 November 2019

ER -

ID: 69222180