JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba

Xiaoyong Lu, Songlin Du*,
Southeast University
CVPR'2025

*Indicates Corresponding Author

Abstract

Existing state-of-the-art feature matchers capture long-range dependencies with Transformers but are hindered by high spatial complexity, leading to demanding training and high-latency inference. Striking a better balance between performance and efficiency remains a challenge in feature matching. Inspired by the linear complexity O(N) of Mamba, we propose an ultra-lightweight Mamba-based matcher, named JamMa, which converges on a single GPU and achieves an impressive performance-efficiency balance in inference. To unlock the potential of Mamba for feature matching, we propose Joint Mamba with a scan-merge strategy named JEGO, which enables: (1) Joint scan of two images to achieve high-frequency mutual interaction, (2) Efficient scan with skip steps to reduce sequence length, (3) Global receptive field, and (4) Omnidirectional feature representation. With the above properties, the JEGO strategy significantly outperforms the scan-merge strategies proposed in VMamba and EVMamba in the feature matching task. Compared to attention-based sparse and semi-dense matchers, JamMa demonstrates a superior balance between performance and efficiency, delivering better performance with less than 50% of the parameters and FLOPs.

JamMa Framework

Descriptive text for the image

JamMa extracts coarse and fine local features with a CNN encoder and scans the coarse features with the JEGO scan module. The four sequences are processed by four independent Mamba blocks and then merged back into 2D feature maps by the JEGO merge module. Finally, the coarse-to-fine matching module (C2F) generates the matching results. The JEGO strategy enables Joint, Efficient, Global, and Omnidirectional scanning and merging.

Detailed presentation of the JamMa pipeline

Effective Receptive Field

Descriptive text for the image

Red circles denote query points, and the ERFs of query points in the two images are concatenated horizontally. JamMa achieves a global and omnidirectional receptive field, a capability made possible through the aggregator.

Results

In all three commonly used efficiency metrics, the proposed JamMa achieves a good performance-efficiency balance.

Descriptive text for the image

Comparison of Qualitative Results.

BibTeX


      @misc{lu2025jamma,
        title={JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba}, 
        author={Xiaoyong Lu and Songlin Du},
        year={2025},
        eprint={2503.03437},
        archivePrefix={arXiv},
        url={https://arxiv.org/abs/2503.03437}, 
  }