Loss it right: Euclidean and Riemannian Metrics in Learning-based Visual Odometry

Olaya Álvarez-Tuñón, Yury Brodskiy, Erdal Kayacan
12/19/2023
cs.CV

Official code: remaro-network/Loss_VO_right (6 stars, MIT, last commit 2024-05-09).

Abstract

This paper overviews different pose representations and metric functions in visual odometry (VO) networks. The performance of VO networks heavily relies on how their architecture encodes the information. The choice of pose representation and loss function significantly impacts network convergence and generalization. We investigate these factors in the VO network DeepVO by implementing loss functions based on Euler, quaternion, and chordal distance and analyzing their influence on performance. The results of this study provide insights into how loss functions affect the designing of efficient and accurate VO networks for camera motion estimation. The experiments illustrate that a distance that complies with the mathematical requirements of a metric, such as the chordal distance, provides better generalization and faster convergence. The code for the experiments can be found at https://github.com/remaro-network/Loss_VO_right

View on arXivView PDF

Code Implementations(1)

remaro-network/Loss_VO_rightOfficial100%
MIT

Comparing the performance of the DeepVO network under different loss functions

60Shell, Python2 years agoMIT
deepvoeuclidean-distancesriemannian-geometryvisual-localisationvisual-localization+2 more

Similar Papers

Brain-Inspired Visual Odometry: Balancing Speed and Interpretability through a System of Systems Approach

87% match1 repo32023

Motion Consistency Loss for Monocular Visual Odometry with Attention-Based Deep Learning

85% match2024

FVO: Fast Visual Odometry with Transformers

84% match2025

Robot Localization and Mapping Final Report -- Sequential Adversarial Learning for Self-Supervised Deep Visual Odometry

84% match2023

From Variance to Veracity: Unbundling and Mitigating Gradient Variance in Differentiable Bundle Adjustment Layers

84% match1 repo212024

MGS-SLAM: Monocular Sparse Tracking and Gaussian Mapping with Depth Smooth Regularization

83% match2024

Cite this paper

@article{lvareztun2023loss,
  title  = {Loss it right: Euclidean and Riemannian Metrics in Learning-based Visual Odometry},
  author = {Olaya Álvarez-Tuñón and Yury Brodskiy and Erdal Kayacan},
  year   = {2023},
  eprint = {2401.05396},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2401.05396v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2401.05396.svg)](https://codeofpaper.com/paper/2401.05396)

Discussion