GatedFusion-Net: a systematic approach to lightweight and efficient RGB–infrared object detection

Fusing Red–Green–Blue (RGB) and infrared images is crucial for all-weather object detection, yet existing high-performance models are often too computationally expensive for deployment on resource-constrained platforms such as drones. To address this, we propose GatedFusion-Net, an end-to-end, lightweight, and efficient detection architecture centered on a systematic lightweight design philosophy. The architecture is composed of our multi-scale module (MS-C3), a symmetric dual-branch backbone, and multi-level GatedFusion layers. Evaluated on several benchmarks, including LLVIP, KAIST, and DroneVehicle, GatedFusion-Net achieves detection accuracy comparable to state-of-the-art models while using only 1.38M parameters and 3.8 Giga Floating-point Operations Per Second (GFLOPs)—a fraction of their computational cost. Our work establishes a new Pareto optimal point between accuracy and efficiency in multi-modal detection, delivering a thoroughly validated and highly efficient solution for practical applications on edge devices.