GatedFusion-Net: A Lightweight and Efficient Network for RGB-Infrared Object Detection

High-performance multi-modal (RGB-infrared) object detectors often suffer from high computational costs, making their real-time deployment on resource-constrained devices challenging. To address this, we propose GatedFusion-Net, a lightweight and efficient network designed to bridge the gap between detection accuracy and operational efficiency. Our core innovation is the GatedFusion module, an attention-based unit that adaptively weights and fuses multi-scale features from RGB and infrared data streams. Experiments on the public LLVIP dataset show that GatedFusion-Net, with only $\mathbf{1. 3 8 M}$ parameters and 3.8 GFLOPs, achieves performance comparable to state-of-the-art (SOTA) models that require over four times the computational cost. GatedFusion-Net provides a practical and viable solution for real-time multi-modal object detection.