LASFNet: A Lightweight Attention-Guided Self-Modulation Feature Fusion Network for Multimodal Object Detection.

Area of Science:

• Computer Vision
• Deep Learning
• Machine Learning

Background:

• Multimodal object detection requires effective deep feature extraction.
• Previous methods face high computational overhead due to complex fusion strategies.

Purpose of the Study:

• To propose a lightweight network for efficient and accurate multimodal object detection.
• To simplify the training process by reducing the number of fusion units.

Main Methods:

• Introduced the lightweight attention-guided self-modulation feature fusion network (LASFNet).
• Employed a single feature-level fusion unit with an attention-guided self-modulation feature fusion (ASFF) module.
• Integrated a feature attention transformation module (FATM) to enhance feature focus.

Main Results:

• LASFNet achieves a favorable efficiency-accuracy tradeoff.
• Reduced parameters by up to 90% and computational cost by 85% compared to state-of-the-art methods.
• Improved mean average precision (mAP) by 1%-3%.

Conclusions:

• LASFNet offers a computationally efficient solution for multimodal object detection.
• The proposed network design enables high-performance detection with reduced complexity.
• The approach demonstrates significant improvements in both efficiency and accuracy.