Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera

© 2019 IEEE. Depth completion, the technique of estimating a dense depth image from sparse depth measurements, has a variety of applications in robotics and autonomous driving. However, depth completion faces 3 main challenges: the irregularly spaced pattern in the sparse depth input, the difficulty...

Full description

Bibliographic Details
Main Authors: Ma, Fangchang, Venturelli Cavalheiro, Guilherme., Karaman, Sertac
Other Authors: Massachusetts Institute of Technology. Department of Aeronautics and Astronautics
Format: Article
Language:English
Published: IEEE 2020
Online Access:https://hdl.handle.net/1721.1/126545
_version_ 1826202847775031296
author Ma, Fangchang
Venturelli Cavalheiro, Guilherme.
Karaman, Sertac
author2 Massachusetts Institute of Technology. Department of Aeronautics and Astronautics
author_facet Massachusetts Institute of Technology. Department of Aeronautics and Astronautics
Ma, Fangchang
Venturelli Cavalheiro, Guilherme.
Karaman, Sertac
author_sort Ma, Fangchang
collection MIT
description © 2019 IEEE. Depth completion, the technique of estimating a dense depth image from sparse depth measurements, has a variety of applications in robotics and autonomous driving. However, depth completion faces 3 main challenges: the irregularly spaced pattern in the sparse depth input, the difficulty in handling multiple sensor modalities (when color images are available), as well as the lack of dense, pixel-level ground truth depth labels for training. In this work, we address all these challenges. Specifically, we develop a deep regression model to learn a direct mapping from sparse depth (and color images) input to dense depth prediction. We also propose a self-supervised training framework that requires only sequences of color and sparse depth images, without the need for dense depth labels. Our experiments demonstrate that the self-supervised framework outperforms a number of existing solutions trained with semi-dense annotations. Furthermore, when trained with semi-dense annotations, our network attains state-of-the-art accuracy and is the winning approach on the KITTI depth completion benchmark² at the time of submission. Furthermore, the self-supervised framework outperforms a number of existing solutions trained with semi-dense annotations.
first_indexed 2024-09-23T12:21:52Z
format Article
id mit-1721.1/126545
institution Massachusetts Institute of Technology
language English
last_indexed 2024-09-23T12:21:52Z
publishDate 2020
publisher IEEE
record_format dspace
spelling mit-1721.1/1265452022-09-28T07:52:47Z Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera Ma, Fangchang Venturelli Cavalheiro, Guilherme. Karaman, Sertac Massachusetts Institute of Technology. Department of Aeronautics and Astronautics © 2019 IEEE. Depth completion, the technique of estimating a dense depth image from sparse depth measurements, has a variety of applications in robotics and autonomous driving. However, depth completion faces 3 main challenges: the irregularly spaced pattern in the sparse depth input, the difficulty in handling multiple sensor modalities (when color images are available), as well as the lack of dense, pixel-level ground truth depth labels for training. In this work, we address all these challenges. Specifically, we develop a deep regression model to learn a direct mapping from sparse depth (and color images) input to dense depth prediction. We also propose a self-supervised training framework that requires only sequences of color and sparse depth images, without the need for dense depth labels. Our experiments demonstrate that the self-supervised framework outperforms a number of existing solutions trained with semi-dense annotations. Furthermore, when trained with semi-dense annotations, our network attains state-of-the-art accuracy and is the winning approach on the KITTI depth completion benchmark² at the time of submission. Furthermore, the self-supervised framework outperforms a number of existing solutions trained with semi-dense annotations. United States. Office of Naval Research (Grant N00014-17-1-2670) 2020-08-12T17:05:50Z 2020-08-12T17:05:50Z 2019-05 2019-10-29T16:03:10Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/126545 Ma, Fangchang, Guilherme Venturelli Cavalheiro and Sertac Karaman. “Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera.” Paper presented at the 2019 International Conference on Robotics and Automation (ICRA), Montreal, QC, Canada, 20-24 May 2019, IEEE © 2019 The Author(s) en 10.1109/ICRA.2019.8793637 2019 International Conference on Robotics and Automation (ICRA) Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf IEEE arXiv
spellingShingle Ma, Fangchang
Venturelli Cavalheiro, Guilherme.
Karaman, Sertac
Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera
title Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera
title_full Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera
title_fullStr Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera
title_full_unstemmed Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera
title_short Self-supervised sparse-to-dense: Self-supervised depth completion from LiDAR and monocular camera
title_sort self supervised sparse to dense self supervised depth completion from lidar and monocular camera
url https://hdl.handle.net/1721.1/126545
work_keys_str_mv AT mafangchang selfsupervisedsparsetodenseselfsuperviseddepthcompletionfromlidarandmonocularcamera
AT venturellicavalheiroguilherme selfsupervisedsparsetodenseselfsuperviseddepthcompletionfromlidarandmonocularcamera
AT karamansertac selfsupervisedsparsetodenseselfsuperviseddepthcompletionfromlidarandmonocularcamera