Multi-Scale Enhanced Depth Knowledge Distillation for Monocular 3D Object Detection with SEFormer

In the context of the Internet of Things, where efficient and accurate perception is crucial. Monocular 3D detection has gained attention due to its cost-effectiveness. This paper introduces an efficient method for monocular 3D object detection, termed Multi-Scale Enhanced Depth Knowledge Distillation (MDKD). Our approach simplifies the teacher network, eliminating the need for extra modal data input while improving the student network's performance. Additionally, we present a Multi-Scale Depth Enhancement (MDE) module and a novel lightweight Squeeze-Excitation Former (SEFormer). Our method addresses the growing demand for precise object detection within IoT environments. Extensive experiments on the KITTI dataset validate our method's effectiveness.