RobustLoc: Robust Camera Pose Regression in Challenging Driving Environments (2022-11-21T00:00:00.000000Z)

TL;DR

This work proposes RobustLoc, which derives its robustness against perturbations from neural differential equations, and uses a convolutional neural network to extract feature maps from multi-view images, and a robust neural differential equation diffusion block module to diffuse information interactively.

Abstract

Camera relocalization has various applications in autonomous driving. Previous camera pose regression models consider only ideal scenarios where there is little environmental perturbation. To deal with challenging driving environments that may have changing seasons, weather, illumination, and the presence of unstable objects, we propose RobustLoc, which derives its robustness against perturbations from neural differential equations. Our model uses a convolutional neural network to extract feature maps from multi-view images, a robust neural differential equation diffusion block module to diffuse information interactively, and a branched pose decoder with multi-layer training to estimate the vehicle poses. Experiments demonstrate that RobustLoc surpasses current state-of-the-art camera pose regression models and achieves robust performance in various environments. Our code is released at: https://github.com/sijieaaa/RobustLoc

Authors

Sijie Wang

2 papers

Qiyu Kang

2 papers

Rui She

2 papers

TL;DR

Abstract

Authors

References33 items

HypLiLoc: Towards Effective LiDAR Pose Regression with Hyperbolic Fusion

On the Robustness of Graph Neural Diffusion to Topology Perturbations

Channel-Wise Attention-Based Network for Self-Supervised Monocular Depth Estimation

Stable Neural ODE with Lyapunov-Stable Equilibrium Points for Defending Against Adversarial Attacks

Beltrami Flow and Neural Diffusion on Graphs

GRAND: Graph Neural Diffusion

CoordiNet: uncertainty-aware pose regressor for reliable vehicle localization

Back to the Feature: Learning Robust Camera Localization from Pixels to Pose

Do We Really Need Scene-specific Pose Encoders?

4Seasons: A Cross-Season Dataset for Multi-Weather SLAM in Autonomous Driving

Learning Multi-View Camera Relocalization With Graph Neural Networks

PointLoc: Deep Pose Regressor for LiDAR Point Cloud Localization

SuperGlue: Learning Feature Matching With Graph Neural Networks

On Robustness of Neural Ordinary Differential Equations

Prior Guided Dropout for Robust Visual Localization in Dynamic Environments

AtLoc: Attention Guided Camera Localization

Local Supports Global: Deep Camera Relocalization With Sequence Enhancement

An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution

Neural Ordinary Differential Equations

Geometry-Aware Learning of Maps for Camera Localization

FC^4: Fully Convolutional Color Constancy with Confidence-Weighted Pooling

Geometric Loss Functions for Camera Pose Regression with Deep Learning

1 year, 1000 km: The Oxford RobotCar dataset

Image-Based Localization Using LSTMs for Structured Feature Correlation

NetVLAD: CNN Architecture for Weakly Supervised Place Recognition

PoseNet: A Convolutional Network for Real-Time 6-DOF Camera Relocalization

Stability and Approximations of Symmetric Diffusion Semigroups and Kernels

Do We Really Need Scenespecific Pose Encoders? In 2020 25th International Conference on Pattern Recognition (ICPR), 3186–3192

ADPoseNet/ADMapNet (Huang et al. 2019), IRPNet(Shavit and Ferens 2021)

After feature map diffusion, a global average pooling module aggregates the feature maps as vector embed-dings, which contain global representations of these images

Codebase Our codes are developed based on the following repositories

Given N neighboring images, a CNN extracts the feature maps of all these images

Table 8: The 4Seasons configuration

Field of Study

Venue Information

Name

Type

URL

Alternate Names