Lightweight Monocular Depth with a Novel Neural Architecture Search Method

Lam Huynh; Phong Nguyen; Jiri Matas; Esa Rahtu; Janne Heikkila

doi:10.1109/WACV51458.2022.00040

Lightweight Monocular Depth with a Novel Neural Architecture Search Method

Lam Huynh, Phong Nguyen, Jiri Matas, Esa Rahtu, Janne Heikkila

Computing Sciences

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

Abstract

This paper presents a novel neural architecture search method, called LiDNAS, for generating lightweight monocular depth estimation models. Unlike previous neural architecture search (NAS) approaches, where finding optimized networks is computationally demanding, the introduced novel Assisted Tabu Search leads to efficient architecture exploration. Moreover, we construct the search space on a pre-defined backbone network to balance layer diversity and search space size. The LiDNAS method outperforms the state-of-the-art NAS approach, proposed for disparity and depth estimation, in terms of search efficiency and output model performance. The LiDNAS optimized models achieve result superior to compact depth estimation state-of-the-art on NYU-Depth-v2, KITTI, and ScanNet, while being 7%-500% more compact in size, i.e the number of model parameters.

Original language	English
Title of host publication	Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022
Publisher	IEEE
Pages	326-336
Number of pages	11
ISBN (Electronic)	9781665409155
DOIs	https://doi.org/10.1109/WACV51458.2022.00040
Publication status	Published - 2022
Publication type	A4 Article in conference proceedings
Event	IEEE/CVF Winter Conference on Applications of Computer Vision - Waikoloa, United States Duration: 4 Jan 2022 → 8 Jan 2022

Publication series

Name	Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022
ISSN (Electronic)	2642-9381

Workshop

Workshop	IEEE/CVF Winter Conference on Applications of Computer Vision
Country/Territory	United States
City	Waikoloa
Period	4/01/22 → 8/01/22

Keywords

3D Computer Vision Deep Learning

Publication forum classification

Publication forum level 1

ASJC Scopus subject areas

Computer Vision and Pattern Recognition
Computer Science Applications

Access to Document

10.1109/WACV51458.2022.00040

2108.11105
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Accepted author manuscript, 8.02 MBLicence: Unspecified

https://urn.fi/URN:NBN:fi:tuni-202211078205Licence: Unspecified

Cite this

Huynh, L., Nguyen, P., Matas, J., Rahtu, E., & Heikkila, J. (2022). Lightweight Monocular Depth with a Novel Neural Architecture Search Method. In Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022 (pp. 326-336). (Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022). IEEE. https://doi.org/10.1109/WACV51458.2022.00040

@inproceedings{1a3ef2fe3cd44c37b77b290f8c9d9af8,

title = "Lightweight Monocular Depth with a Novel Neural Architecture Search Method",

abstract = "This paper presents a novel neural architecture search method, called LiDNAS, for generating lightweight monocular depth estimation models. Unlike previous neural architecture search (NAS) approaches, where finding optimized networks is computationally demanding, the introduced novel Assisted Tabu Search leads to efficient architecture exploration. Moreover, we construct the search space on a pre-defined backbone network to balance layer diversity and search space size. The LiDNAS method outperforms the state-of-the-art NAS approach, proposed for disparity and depth estimation, in terms of search efficiency and output model performance. The LiDNAS optimized models achieve result superior to compact depth estimation state-of-the-art on NYU-Depth-v2, KITTI, and ScanNet, while being 7%-500% more compact in size, i.e the number of model parameters. ",

keywords = "3D Computer Vision Deep Learning",

author = "Lam Huynh and Phong Nguyen and Jiri Matas and Esa Rahtu and Janne Heikkila",

note = "Publisher Copyright: {\textcopyright} 2022 IEEE.; IEEE/CVF Winter Conference on Applications of Computer Vision ; Conference date: 04-01-2022 Through 08-01-2022",

year = "2022",

doi = "10.1109/WACV51458.2022.00040",

language = "English",

series = "Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022",

publisher = "IEEE",

pages = "326--336",

booktitle = "Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022",

}

Huynh, L, Nguyen, P, Matas, J, Rahtu, E & Heikkila, J 2022, Lightweight Monocular Depth with a Novel Neural Architecture Search Method. in Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022. Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022, IEEE, pp. 326-336, IEEE/CVF Winter Conference on Applications of Computer Vision, Waikoloa, United States, 4/01/22. https://doi.org/10.1109/WACV51458.2022.00040

Lightweight Monocular Depth with a Novel Neural Architecture Search Method. / Huynh, Lam; Nguyen, Phong; Matas, Jiri et al.
Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022. IEEE, 2022. p. 326-336 (Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Lightweight Monocular Depth with a Novel Neural Architecture Search Method

AU - Huynh, Lam

AU - Nguyen, Phong

AU - Matas, Jiri

AU - Rahtu, Esa

AU - Heikkila, Janne

PY - 2022

Y1 - 2022

N2 - This paper presents a novel neural architecture search method, called LiDNAS, for generating lightweight monocular depth estimation models. Unlike previous neural architecture search (NAS) approaches, where finding optimized networks is computationally demanding, the introduced novel Assisted Tabu Search leads to efficient architecture exploration. Moreover, we construct the search space on a pre-defined backbone network to balance layer diversity and search space size. The LiDNAS method outperforms the state-of-the-art NAS approach, proposed for disparity and depth estimation, in terms of search efficiency and output model performance. The LiDNAS optimized models achieve result superior to compact depth estimation state-of-the-art on NYU-Depth-v2, KITTI, and ScanNet, while being 7%-500% more compact in size, i.e the number of model parameters.

AB - This paper presents a novel neural architecture search method, called LiDNAS, for generating lightweight monocular depth estimation models. Unlike previous neural architecture search (NAS) approaches, where finding optimized networks is computationally demanding, the introduced novel Assisted Tabu Search leads to efficient architecture exploration. Moreover, we construct the search space on a pre-defined backbone network to balance layer diversity and search space size. The LiDNAS method outperforms the state-of-the-art NAS approach, proposed for disparity and depth estimation, in terms of search efficiency and output model performance. The LiDNAS optimized models achieve result superior to compact depth estimation state-of-the-art on NYU-Depth-v2, KITTI, and ScanNet, while being 7%-500% more compact in size, i.e the number of model parameters.

KW - 3D Computer Vision Deep Learning

U2 - 10.1109/WACV51458.2022.00040

DO - 10.1109/WACV51458.2022.00040

M3 - Conference contribution

AN - SCOPUS:85126147160

T3 - Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022

SP - 326

EP - 336

BT - Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022

PB - IEEE

T2 - IEEE/CVF Winter Conference on Applications of Computer Vision

Y2 - 4 January 2022 through 8 January 2022

ER -

Huynh L, Nguyen P, Matas J, Rahtu E, Heikkila J. Lightweight Monocular Depth with a Novel Neural Architecture Search Method. In Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022. IEEE. 2022. p. 326-336. (Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022). doi: 10.1109/WACV51458.2022.00040

Lightweight Monocular Depth with a Novel Neural Architecture Search Method

Abstract

Publication series

Workshop

Keywords

Publication forum classification

ASJC Scopus subject areas

Access to Document

Fingerprint

Cite this