MambaFPN: A SSM-based feature pyramid network for object detection.

Area of Science:

• Computer Vision
• Deep Learning

Background:

• Object detection is vital in computer vision.
• Feature Pyramid Networks (FPNs) are crucial for handling objects of varying sizes.
• Convolutional Neural Networks (CNNs) in FPNs limit global context capture.

Purpose of the Study:

• To enhance global modeling capabilities in FPNs.
• To address the limitations of CNNs in capturing global context.
• To improve object detection performance and efficiency.

Main Methods:

• Leveraging vision mamba blocks for global context modeling.
• Utilizing vision mamba blocks to extract global information from individual feature maps.
• Employing additional vision mamba blocks for inter-scale information exchange.

Main Results:

• The proposed MambaFPN significantly improves object detector performance.
• MambaFPN enhanced the Average Precision (AP) of vanilla FPN from 38.6 to 39.4.
• The method achieves these improvements with fewer parameters, demonstrating efficiency.

Conclusions:

• MambaFPN effectively integrates global context into object detection.
• The use of vision mamba blocks offers a promising direction for advancing FPNs.
• MambaFPN presents an effective and efficient solution for object detection tasks.