Structure-Centric Robust Monocular Depth Estimation via Knowledge Distillation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Runze, Luo, Haiyong, Zhao, Fang, Yu, Jingze, Jia, Yupeng, Wang, Juan, Ma, Xuepeng
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913539368681472
author Chen, Runze
Luo, Haiyong
Zhao, Fang
Yu, Jingze
Jia, Yupeng
Wang, Juan
Ma, Xuepeng
author_facet Chen, Runze
Luo, Haiyong
Zhao, Fang
Yu, Jingze
Jia, Yupeng
Wang, Juan
Ma, Xuepeng
contents Monocular depth estimation, enabled by self-supervised learning, is a key technique for 3D perception in computer vision. However, it faces significant challenges in real-world scenarios, which encompass adverse weather variations, motion blur, as well as scenes with poor lighting conditions at night. Our research reveals that we can divide monocular depth estimation into three sub-problems: depth structure consistency, local texture disambiguation, and semantic-structural correlation. Our approach tackles the non-robustness of existing self-supervised monocular depth estimation models to interference textures by adopting a structure-centered perspective and utilizing the scene structure characteristics demonstrated by semantics and illumination. We devise a novel approach to reduce over-reliance on local textures, enhancing robustness against missing or interfering patterns. Additionally, we incorporate a semantic expert model as the teacher and construct inter-model feature dependencies via learnable isomorphic graphs to enable aggregation of semantic structural knowledge. Our approach achieves state-of-the-art out-of-distribution monocular depth estimation performance across a range of public adverse scenario datasets. It demonstrates notable scalability and compatibility, without necessitating extensive model engineering. This showcases the potential for customizing models for diverse industrial applications.
format Preprint
id arxiv_https___arxiv_org_abs_2410_06982
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Structure-Centric Robust Monocular Depth Estimation via Knowledge Distillation
Chen, Runze
Luo, Haiyong
Zhao, Fang
Yu, Jingze
Jia, Yupeng
Wang, Juan
Ma, Xuepeng
Computer Vision and Pattern Recognition
Monocular depth estimation, enabled by self-supervised learning, is a key technique for 3D perception in computer vision. However, it faces significant challenges in real-world scenarios, which encompass adverse weather variations, motion blur, as well as scenes with poor lighting conditions at night. Our research reveals that we can divide monocular depth estimation into three sub-problems: depth structure consistency, local texture disambiguation, and semantic-structural correlation. Our approach tackles the non-robustness of existing self-supervised monocular depth estimation models to interference textures by adopting a structure-centered perspective and utilizing the scene structure characteristics demonstrated by semantics and illumination. We devise a novel approach to reduce over-reliance on local textures, enhancing robustness against missing or interfering patterns. Additionally, we incorporate a semantic expert model as the teacher and construct inter-model feature dependencies via learnable isomorphic graphs to enable aggregation of semantic structural knowledge. Our approach achieves state-of-the-art out-of-distribution monocular depth estimation performance across a range of public adverse scenario datasets. It demonstrates notable scalability and compatibility, without necessitating extensive model engineering. This showcases the potential for customizing models for diverse industrial applications.
title Structure-Centric Robust Monocular Depth Estimation via Knowledge Distillation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.06982