Title

A Study on the Generality of a DNN-Based Monocular Depth Estimation

DGIST Authors

Jinwoo Bae ; Sunghoon Im ; Kyong Hwan Jin

Advisor

임성훈

Co-Advisor(s)

Kyong Hwan Jin

Issued Date

2023

Awarded Date

2023-02-01

Citation

Jinwoo Bae. (2023). A Study on the Generality of a DNN-Based Monocular Depth Estimation. doi: 10.22677/THESIS.200000653335

Type

Thesis

Description

Monocular depth estimation, Generalization, Out-of-Distribution

Table Of Contents

Ⅰ. Introduction 1
Ⅱ. Related Work 2
2.1 Self-supervised monocular depth estimation 2
2.2 Supervised monocular depth estimation 2
2.3 Vision Transformers 3
2.4 Modernized Architecture 4
Ⅲ. Method 4
3.1 CNN-Transformer Encoder 5
3.2 Attention Connection Module 6
3.3 Feature Fusion Decoder 7
Ⅳ. Experimental Results 7
4.1 Comparison on KITTI datasets 8
4.2 Analysis of texture-/shape-bias on CNN and Transformer 9
4.3 Analysis of Generalization performance 18
4.4 Analysis of feature representation on CNN and Transformer 21
Ⅴ. Conclusion 23
Ⅵ. References 25
Ⅶ. 요약문 31

URI

http://hdl.handle.net/20.500.11750/45756
http://dgist.dcollection.net/common/orgView/200000653335

DOI

10.22677/THESIS.200000653335

Degree

Master

Department

Department of Electrical Engineering and Computer Science

Publisher

DGIST

Show Full Item Record