EA-EDNet: encapsulated attention encoder-decoder network for 3D reconstruction in low-light-level environment

3D reconstruction via neural networks has become striking nowadays. However, the existing works are based on information-rich environment to perform reconstruction, not yet about the Low-Light-Level (LLL) environment where the information is extremely scarce. The implementation of 3D reconstruction...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia systems Vol. 29; no. 4; pp. 2263 - 2279
Main Authors: Deng, Yulin, Yin, Liju, Gao, Xiaoning, Zhou, Hui, Wang, Zhenzhou, Zou, Guofeng
Format: Journal Article
Language:English
Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2023
Springer Nature B.V
Subjects:
Computer Communication Networks
Computer Graphics
Computer Science
Cryptology
Data Storage Representation
Datasets
Deep learning
Encapsulation
Encoders-Decoders
Image reconstruction
Morphology
Multimedia Information Systems
Neural networks
Operating Systems
Regular Paper
Training
Computer stereo vision
Low-light-level environment imaging
3D reconstruction
ISSN:0942-4962, 1432-1882
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:3D reconstruction via neural networks has become striking nowadays. However, the existing works are based on information-rich environment to perform reconstruction, not yet about the Low-Light-Level (LLL) environment where the information is extremely scarce. The implementation of 3D reconstruction in this environment is an urgent requirement for military, aerospace and other fields. Therefore, we introduce an Encapsulated Attention Encoder-Decoder Network (EA-EDNet) in this paper. It can incorporate multiple levels of semantic to adequately extract the limited information from images taken in the LLL environment and can reason out the defective morphological data as well as intensify the attention to the focused parts. The EA-EDNet adopts a two-stage combined coarse-to-fine training fashion. We additionally create a realistic LLL environment dataset 3LNet-12, and accompanying propose an analysis method for filtering this dataset. In experiments, the proposed method not only achieves results superior to the state-of-the-art methods, but also achieves more delicate reconstruction models.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0942-4962
1432-1882
DOI:10.1007/s00530-023-01100-2