Shuffle in 3D: A Lightweight Architecture for Stereo Matching

The deep learning-based stereo matching approaches commonly construct 3D cost volume with a Siamese network, and the 3D encoder-decoder architectures regularize 3D cost volume to generate disparity as output. Yet, the 3D encoder-decoder architectures produce tremendous parameters that causing huge t...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2021 IEEE 10th Global Conference on Consumer Electronics (GCCE) S. 675 - 677
Hauptverfasser: Xiao, Jianqiang, Yamane, Satoshi
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 12.10.2021
Schlagworte:
3D ShuffleNet
Computational efficiency
Computer architecture
Conferences
Consumer electronics
Costs
lightweight 3D encoder-decoder
Stereo Matching
Three-dimensional displays
Training
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The deep learning-based stereo matching approaches commonly construct 3D cost volume with a Siamese network, and the 3D encoder-decoder architectures regularize 3D cost volume to generate disparity as output. Yet, the 3D encoder-decoder architectures produce tremendous parameters that causing huge training and inference cost. In this paper, we propose a 3D ShuffleNet-based approach on PSMNet, which greatly saves the computational cost without losing without reducing the performance of the network.
DOI:10.1109/GCCE53005.2021.9622083