English

2025 Project

SFMambaNet: Spectral-Frequency Correspondence Pruning

A spectral-frequency enhanced selective state space model for efficient visual correspondence pruning.

Computer VisionState Space ModelsMambaCorrespondence PruningFrequency DomainGraph Neural NetworksTwo-View GeometryCamera Pose EstimationOutlier Rejection
arXiv / DOIhttps://doi.org/10.48550/arXiv.2606.04493 IEEE TIPhttps://signalprocessingsociety.org/publications-resources/ieee-transactions-image-processing

SFMambaNet investigates efficient global context modeling for correspondence pruning.

Paper labels: First author; SCI Zone 1; CCF A; under review at IEEE Transactions on Image Processing (TIP).

SFMambaNet framework

The project introduces spectral-frequency awareness into a selective state space architecture. This direction aims to avoid the quadratic scaling of Transformer-style global attention while keeping strong global modeling capacity.

This work was conducted during a research internship at the Fujian Key Laboratory of Network Computing and Intelligent Information Processing, Fuzhou University, under the supervision of Advisor Yizhang Liu. The manuscript is currently under review at IEEE Transactions on Image Processing (TIP) and remains available as an arXiv preprint.