View in EDS

HMSNet: Hilbert curve enhanced Mamba for real-time semantic segmentation.

Saved in:
Bibliographic Details
Title: HMSNet: Hilbert curve enhanced Mamba for real-time semantic segmentation.
Authors: Jia, Lianyin1,2 (AUTHOR) lianyinjia@kust.edu.cn, Gao, Aoxiang1 (AUTHOR) axgao@stu.kust.edu.cn, Li, Mengjuan3 (AUTHOR) lmjlykm@163.com, Fu, Xiaodong1 (AUTHOR) xiaodong_fu@hotmail.com, Zhou, Haihe1 (AUTHOR) 18908715777@189.cn, Ding, Jiaman1 (AUTHOR) jiamanding@kust.edu.cn
Source: Pattern Recognition. Apr2026:Part B, Vol. 172, pN.PAG-N.PAG. 1p.
Subjects: Image segmentation, Real-time computing, Hilbert transform, Context-aware computing, Software frameworks
Abstract: • Propose HMSNet, a novel real-time semantic segmentation architecture based on a single-branch network, balancing speed and accuracy. • Introduce the HSS2D scanning technique, combining the Hilbert curve with Mamba to construct the HVM Block, alleviating intra-class semantic inconsistency. • Propose the Selective Attention Fusion Module (SAFM), integrating shallow detail features with deep semantic features to mitigate local detail loss. • Design the Multi-scale Context-Aware Module (MCAM), enhancing multi-scale semantic information to improve segmentation accuracy. • HMSNet outperforms existing methods in both segmentation accuracy and inference speed on CamVid, Cityscapes and ADE20K. Semantic segmentation is a core technology for vehicle perception of the surrounding environment in autonomous driving. However, existing real-time semantic segmentation models face two major challenges: loss of local detail information and inconsistency of intra-class semantic information. To address these issues, we propose a novel network architecture, HMSNet. The network mainly consists of the following three core modules: the Hilbert curve enhanced Visual Mamba Block (HVM Block), Selective Attention Fusion Module (SAFM), and Multi-scale Context-Aware Module (MCAM). The HVM Block utilizes the Hilbert curve to reduce the dimensionality of two-dimensional images and applies a selective scanning algorithm in Mamba, enabling the network to effectively capture local dependencies while maintaining a global receptive field, thereby optimizing the consistency of intra-class semantic information. The SAFM module effectively merges local detail information from shallow networks with global semantic information from deep networks, alleviating the problem of local detail information loss. Finally, the MCAM module, introduced at the end of the network, enhances the model,s ability to judge contextual information, thereby improving segmentation accuracy. Experimental results show that HMSNet achieves an excellent balance between segmentation accuracy and inference speed on challenging public datasets, including CamVid, Cityscapes, and ADE20K. [ABSTRACT FROM AUTHOR]
Copyright of Pattern Recognition is the property of Pergamon Press - An Imprint of Elsevier Science and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:• Propose HMSNet, a novel real-time semantic segmentation architecture based on a single-branch network, balancing speed and accuracy. • Introduce the HSS2D scanning technique, combining the Hilbert curve with Mamba to construct the HVM Block, alleviating intra-class semantic inconsistency. • Propose the Selective Attention Fusion Module (SAFM), integrating shallow detail features with deep semantic features to mitigate local detail loss. • Design the Multi-scale Context-Aware Module (MCAM), enhancing multi-scale semantic information to improve segmentation accuracy. • HMSNet outperforms existing methods in both segmentation accuracy and inference speed on CamVid, Cityscapes and ADE20K. Semantic segmentation is a core technology for vehicle perception of the surrounding environment in autonomous driving. However, existing real-time semantic segmentation models face two major challenges: loss of local detail information and inconsistency of intra-class semantic information. To address these issues, we propose a novel network architecture, HMSNet. The network mainly consists of the following three core modules: the Hilbert curve enhanced Visual Mamba Block (HVM Block), Selective Attention Fusion Module (SAFM), and Multi-scale Context-Aware Module (MCAM). The HVM Block utilizes the Hilbert curve to reduce the dimensionality of two-dimensional images and applies a selective scanning algorithm in Mamba, enabling the network to effectively capture local dependencies while maintaining a global receptive field, thereby optimizing the consistency of intra-class semantic information. The SAFM module effectively merges local detail information from shallow networks with global semantic information from deep networks, alleviating the problem of local detail information loss. Finally, the MCAM module, introduced at the end of the network, enhances the model,s ability to judge contextual information, thereby improving segmentation accuracy. Experimental results show that HMSNet achieves an excellent balance between segmentation accuracy and inference speed on challenging public datasets, including CamVid, Cityscapes, and ADE20K. [ABSTRACT FROM AUTHOR]
ISSN:00313203
DOI:10.1016/j.patcog.2025.112457