Abstract ABSTRACT— Achieving efficient and accurate semantic &amp;#xD;segmentation of LiDAR point clouds is a crucial fundamental &amp;#xD;technology in autonomous driving and robotics. In the paper, &amp;#xD;We have designed the convolutional stem of patch embedding &amp;#xD;before the ViT in order to enhance RangeViT and improve its &amp;#xD;performance in point cloud semantic segmentation. And we &amp;#xD;have named this improved version RangeViT++. Firstly, a &amp;#xD;Multi Residual Channel Interaction Attention Module &amp;#xD;(MRCIAM) is introduced to replace original context module &amp;#xD;of RangeViT, utilizing a multi-branch structure to separately &amp;#xD;process the various channels of the range image in order to &amp;#xD;consider their modality and data distribution differences. &amp;#xD;Secondly, the Meta-Kernel module is introduced to mitigate &amp;#xD;information loss caused by the traditional CNN's incomplete &amp;#xD;adaptation to point cloud range images, which fail to fully &amp;#xD;exploit the inherent 3D geometric information of point clouds. &amp;#xD;Lastly, during the training process, a boundary loss is incor porated to alleviate the boundary ambiguity of different clas ses/objects induced by the mutual conversion between point &amp;#xD;clouds and range images. Extensive qualitative and quantita tive experiments conducted on challenging SemanticKITTI &amp;#xD;and SemanticPOSS dataset have verified effectiveness of our &amp;#xD;method. Superior performance is present over baseline &amp;#xD;RangeViT, which indicates refining the early convolutional &amp;#xD;stem could improve the performance of ViT on LiDAR point &amp;#xD;cloud semantic segmentation. The source code and trained &amp;#xD;model will be available at https://github.com/mafangniu/&amp;#xD;RangeViT2.git

Chunyun Ma

Xiaojun Shi

Lu Chen

Shuai Song

Yingxin Wang

Jiaxiang Hu

Zhen Pan

Physica Scripta

PHYSICA SCRIPTA 是一本国际期刊，针对实验和理论物理学的任何分支进行原创研究。文章将在以下任一主题中被考虑，也欢迎涉及物理学的跨学科主题：-原子、分子和光学物理-等离子体物理-凝聚态物理-数学物理-天体物理-高能物理-核物理-非线性物理。该杂志旨在提高研究对更广泛的物理科学界的知名度和可及性。鼓励发表关于广泛感兴趣的主题的文章，并且在更专业的领域提交的文章应努力在引言中提及他们研究的更广泛背景。

Physica Scripta is an international journal dedicated to original research in all branches of experimental and theoretical physics. We consider articles on any of the following topics and welcome interdisciplinary subjects related to physics: - Atomic, molecular, and optical physics - Plasma physics - Condensed matter physics - Mathematical physics - Astrophysics - High-energy physics - Nuclear physics - Nonlinear physics. The journal aims to enhance the visibility and accessibility of research to the broader physics community. We encourage submissions on topics of wide interest, and articles submitted in more specialized fields should strive to mention the broader context of their research in the introduction.

《Physica Scripta》是一本国际期刊，专注于实验和理论物理学的各个分支，欢迎原创研究的投稿。我们接受以下主题的文章，同时也鼓励涉及物理学的跨学科研究：- 原子、分子和光学物理 - 等离子体物理 - 凝聚态物理 - 数学物理 - 天体物理 - 高能物理 - 核物理 - 非线性物理。该期刊旨在提升研究成果在更广泛物理科学界的知名度和可及性。我们鼓励发表与广泛主题相关的文章，同时在更专业领域的投稿应努力在引言中阐明其研究的更广泛背景。

Category	Quartile
物理与天体物理	3
物理与天体物理, 物理综合	3

RangeViT++: Refining the early convolutional stem help ViT see better

ivySCI AI Smartly Parses PDF, Answers Researchers' Questions, and Helps You Understand Papers in Seconds

Journal Info

Category	Quartile
PHYSICS, MULTIDISCIPLINARY	2