Skip to main navigation Skip to search Skip to main content

APAC-Net: Unsupervised Learning of Depth and Ego-Motion from Monocular Video

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

We propose an unsupervised novel method, Attention-Pixel and Attention-Channel Network (APAC-Net), for unsupervised monocular learning of estimating scene depth and ego-motion. Our model only utilizes monocular image sequences and does not need additional sensor information, such as IMU and GPS, for supervising. The attention mechanism is employed in APAC-Net to improve the networks’ efficiency. Specifically, three attention modules are proposed to adjust feature weights when training. Moreover, to minimum the effect of noise, which is produced in the reconstruction processing, the Image-reconstruction loss based on PSNR LPSNR is used to evaluation the reconstruction quality. In addition, due to the fail depth estimation of the objects closed to camera, the Temporal-consistency loss LTemp between adjacent frames and the Scale-based loss LScale among different scales are proposed. Experimental results showed APAC-Net can perform well in both the depth and ego-motion tasks, and it even behaved better in several items on KITTI and Cityscapes.

Original languageEnglish
Title of host publicationIntelligence Science and Big Data Engineering. Visual Data Engineering - 9th International Conference, IScIDE 2019, Proceedings, Part 1
EditorsZhen Cui, Jinshan Pan, Shanshan Zhang, Liang Xiao, Jian Yang
PublisherSpringer
Pages336-348
Number of pages13
ISBN (Print)9783030361884
DOIs
StatePublished - 2019
Externally publishedYes
Event9th International Conference on Intelligence Science and Big Data Engineering, IScIDE 2019 - Nanjing, China
Duration: 17 Oct 201920 Oct 2019

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11935 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference9th International Conference on Intelligence Science and Big Data Engineering, IScIDE 2019
Country/TerritoryChina
CityNanjing
Period17/10/1920/10/19

Keywords

  • Attention mechanism
  • Depth estimation
  • Ego-motion estimation

Fingerprint

Dive into the research topics of 'APAC-Net: Unsupervised Learning of Depth and Ego-Motion from Monocular Video'. Together they form a unique fingerprint.

Cite this