MambaPSA: A Mamba-based Replacement for C2PSA in YOLO26

2026-07-14

AI Summary

This work addresses the high computational cost of the C2PSA module in the lightweight, NMS-free object detector YOLO26 by introducing the state space model Mamba into its architecture for the first time. Specifically, C2PSA is replaced with MambaPSA at the end of the backbone, and bidirectional Vision Mamba (BiViM) modules are embedded into the P3–P5 layers of the neck. This design achieves a superior trade-off between efficiency and accuracy, reducing parameters by 2.9% and FLOPs by 12.1% on PASCAL VOC, while accelerating CPU inference by 17.6% (from 17 to 20 FPS) with only a marginal 0.1 drop in mAP50:95. Notably, the BiViM module at the P4 layer alone contributes a +0.9 gain in mAP50:95, demonstrating the effectiveness and efficiency of Mamba within NMS-free detection frameworks.

Abstract

State space models (SSMs), notably Mamba, have recently emerged as efficient alternatives to self-attention with linear computational complexity. We investigate the integration of Mamba into YOLO26, the latest non-maximum suppression (NMS)-free object detection framework, by proposing MambaPSA, a lightweight Mamba-based replacement for the C2PSA block at the end of the backbone. To complement this study, we additionally insert a bidirectional Vision Mamba (BiViM) module at the P3, P4, and P5 levels of the neck. Experiments on PASCAL VOC 2007+2012 show that MambaPSA reduces parameters by 2.9%, FLOPs by 12.1%, and improves CPU inference throughput by 17.6% (from 17 to 20 FPS) with negligible accuracy change (-0.1 mAP50:95), while the P4 BiViM placement yields the best accuracy gain (+0.9 mAP50:95). These results suggest that SSMs offer a favorable efficiency-accuracy trade-off when replacing attention-based blocks in NMS-free lightweight detectors.

Problem

• object detection
• state space models
• Mamba
• YOLO
• efficiency-accuracy trade-off

Innovation

• Mamba
• State Space Models
• YOLO26
• C2PSA replacement
• BiViM