Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Parameterization of Ambiguity in Monocular Depth Prediction

Persson, Patrik LU orcid ; Ostrom, Linn LU ; Olsson, Carl LU and Astrom, Kalle LU orcid (2021) 9th International Conference on 3D Vision, 3DV 2021 p.761-770
Abstract

Monocular depth estimation is a highly challenging problem that is often addressed with deep neural networks. While these use recognition of high level image features to predict reasonably looking depth maps,the result often has poor metric accuracy. Moreover,the standard feed forward architecture does not allow modification of the prediction based on cues other than the image.In this paper we relax the monocular depth estimation task by proposing a network that allows us to complement image features with a set of auxiliary variables. These allow disambiguation when image features are not enough to accurately pinpoint the exact depth map and can be thought of as a low dimensional parameterization of the surfaces that are reasonable... (More)

Monocular depth estimation is a highly challenging problem that is often addressed with deep neural networks. While these use recognition of high level image features to predict reasonably looking depth maps,the result often has poor metric accuracy. Moreover,the standard feed forward architecture does not allow modification of the prediction based on cues other than the image.In this paper we relax the monocular depth estimation task by proposing a network that allows us to complement image features with a set of auxiliary variables. These allow disambiguation when image features are not enough to accurately pinpoint the exact depth map and can be thought of as a low dimensional parameterization of the surfaces that are reasonable monocular predictions. By searching the parameterization we can combine monocular estimation with traditional photoconsistency or geometry based methods to achieve both visually appealing and metrically accurate surface estimations. Since we relax the problem we are able to work with smaller networks than current architectures. In addition we design a self-supervised training scheme,eliminating the need for ground truth image depth-map pairs. Our experimental evaluation shows that our method generates more accurate depth maps and generalizes better than competing state-of-the-art approaches.

(Less)
Please use this url to cite or link to this publication:
author
; ; and
organization
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
published
subject
keywords
3D Reconstruction, Machine Learning, Monocular Depth Parameterization
host publication
Proceedings - 2021 International Conference on 3D Vision, 3DV 2021
pages
10 pages
publisher
IEEE - Institute of Electrical and Electronics Engineers Inc.
conference name
9th International Conference on 3D Vision, 3DV 2021
conference location
Virtual, Online, United Kingdom
conference dates
2021-12-01 - 2021-12-03
external identifiers
  • scopus:85125009837
ISBN
9781665426886
DOI
10.1109/3DV53792.2021.00085
language
English
LU publication?
yes
id
215fdfde-0a06-4dba-ad8e-421cde7baa5d
date added to LUP
2022-04-13 17:06:02
date last changed
2022-05-07 00:34:32
@inproceedings{215fdfde-0a06-4dba-ad8e-421cde7baa5d,
  abstract     = {{<p>Monocular depth estimation is a highly challenging problem that is often addressed with deep neural networks. While these use recognition of high level image features to predict reasonably looking depth maps,the result often has poor metric accuracy. Moreover,the standard feed forward architecture does not allow modification of the prediction based on cues other than the image.In this paper we relax the monocular depth estimation task by proposing a network that allows us to complement image features with a set of auxiliary variables. These allow disambiguation when image features are not enough to accurately pinpoint the exact depth map and can be thought of as a low dimensional parameterization of the surfaces that are reasonable monocular predictions. By searching the parameterization we can combine monocular estimation with traditional photoconsistency or geometry based methods to achieve both visually appealing and metrically accurate surface estimations. Since we relax the problem we are able to work with smaller networks than current architectures. In addition we design a self-supervised training scheme,eliminating the need for ground truth image depth-map pairs. Our experimental evaluation shows that our method generates more accurate depth maps and generalizes better than competing state-of-the-art approaches. </p>}},
  author       = {{Persson, Patrik and Ostrom, Linn and Olsson, Carl and Astrom, Kalle}},
  booktitle    = {{Proceedings - 2021 International Conference on 3D Vision, 3DV 2021}},
  isbn         = {{9781665426886}},
  keywords     = {{3D Reconstruction; Machine Learning; Monocular Depth Parameterization}},
  language     = {{eng}},
  pages        = {{761--770}},
  publisher    = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}},
  title        = {{Parameterization of Ambiguity in Monocular Depth Prediction}},
  url          = {{http://dx.doi.org/10.1109/3DV53792.2021.00085}},
  doi          = {{10.1109/3DV53792.2021.00085}},
  year         = {{2021}},
}