City-scale continual neural semantic mapping with three-layer sampling and panoptic representation

Shi, Yongliang; Yang, Runyi; Wu, Zirui; Li, Pengfei; Liu, Caiyun; Zhao, Hao; Zhou, Guyue

doi:10.1016/j.knosys.2023.111145

Cited by 2 publications

(1 citation statement)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A large number of invalid points are generated at the beginning of sampling, which makes network convergence slow. In order to accelerate the training speed, Shi and Yang et al [37] proposed a mapping method based on a three-layer sampling strategy. In addition to global sampling, local sampling is introduced.…”

Section: B Implicit Joint Representations Using Voxelsmentioning

confidence: 99%

SLAM Meets NeRF: A Survey of Implicit SLAM Methods

Yang,

Cheng,

Chen

et al. 2024

WEVJ

View full text Add to dashboard Cite

In recent years, Simultaneous Localization and Mapping (SLAM) systems have shown significant performance, accuracy, and efficiency gains, especially when Neural Radiance Fields (NeRFs) are implemented. NeRF-based SLAM in mapping aims to implicitly understand irregular environmental information using large-scale parameters of deep learning networks in a data-driven manner so that specific environmental information can be predicted from a given perspective. NeRF-based SLAM in tracking jointly optimizes camera pose and implicit scene network parameters through inverse rendering or combines VO and NeRF mapping to achieve real-time positioning and mapping. This paper firstly analyzes the current situation of NeRF and SLAM systems and then introduces the state-of-the-art in NeRF-based SLAM. In addition, datasets and system evaluation methods used by NeRF-based SLAM are introduced. In the end, current issues and future work are analyzed. Based on an investigation of 30 related research articles, this paper provides in-depth insight into the innovation of SLAM and NeRF methods and provides a useful reference for future research.

show abstract

Section: B Implicit Joint Representations Using Voxelsmentioning

confidence: 99%