Scholarlink

<p>Local feature extraction and matching has lately attracted increasing attention due to its wide application, especially in real-time automated systems. However, existing image matching methods struggle to balance the global receptive field and the efficient computation, which limits the practical applications. Recently, the State Space Model (SSM) has shown great potential in linear complexity and long-range dependency modeling. Therefore, in this paper, a local feature extraction and matching method using the SSM is proposed, which aims to achieve the tradeoff between global information extraction and model complexity. Firstly, a Local and Global Information Fusion (LGIF) block is developed to integrate local and global information, and&#xa0;to reduce model parameters through parallel SSM. Secondly, a backbone based on Euclidean group E(2) equivariant steerable Convolution (E2Conv) is designed to improve the model’s robustness against geometric transformations. Finally, a self-supervised learning framework is constructed, which optimizes the ability of the network in local feature detection and description by combining four loss functions: local feature localization loss, local feature confidence score loss, descriptor triplet loss, and local feature correspondence loss. Experimental results on public benchmark datasets Hpatches and RDNIM demonstrate that the proposed method has a significant advantage over existing methods in homography estimation tasks. For example, in the HPatches dataset, under the 1-pixel error threshold in the HA metric, our method outperforms SuperPoint by 8.74% in sparse point matching (up to 500 local features) and surpasses LoFTR by 8.15% in dense point matching (up to 2000 local features). The PyTorch code is available at: <a href="https://github.com/TakeoffC/MambaLF">https://github.com/TakeoffC/MambaLF</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MambaLF: an efficient local feature extraction and matching with state space model

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Houqin Bian,
  • Qifei Chen,
  • Haolin Zhang,
  • Lunming Qin,
  • Liang Xue,
  • Haoyang Cui,
  • Xi Wang

摘要

Local feature extraction and matching has lately attracted increasing attention due to its wide application, especially in real-time automated systems. However, existing image matching methods struggle to balance the global receptive field and the efficient computation, which limits the practical applications. Recently, the State Space Model (SSM) has shown great potential in linear complexity and long-range dependency modeling. Therefore, in this paper, a local feature extraction and matching method using the SSM is proposed, which aims to achieve the tradeoff between global information extraction and model complexity. Firstly, a Local and Global Information Fusion (LGIF) block is developed to integrate local and global information, and to reduce model parameters through parallel SSM. Secondly, a backbone based on Euclidean group E(2) equivariant steerable Convolution (E2Conv) is designed to improve the model’s robustness against geometric transformations. Finally, a self-supervised learning framework is constructed, which optimizes the ability of the network in local feature detection and description by combining four loss functions: local feature localization loss, local feature confidence score loss, descriptor triplet loss, and local feature correspondence loss. Experimental results on public benchmark datasets Hpatches and RDNIM demonstrate that the proposed method has a significant advantage over existing methods in homography estimation tasks. For example, in the HPatches dataset, under the 1-pixel error threshold in the HA metric, our method outperforms SuperPoint by 8.74% in sparse point matching (up to 500 local features) and surpasses LoFTR by 8.15% in dense point matching (up to 2000 local features). The PyTorch code is available at: https://github.com/TakeoffC/MambaLF.