A Transformer-Based End-to-End Network for Unmanned Aerial Vehicle Aerial Image Object Detection

Abstract

The Transformer-based end-to-end networks have received extensive attention from the academia and industry due to their superior detection performance and elimination of handcrafted components. However, their high computational costs have hindered their applicability in unmanned aerial vehicle (UAV) aerial image object detection tasks driven by robotics technology. This paper proposes a novel Transformer-based end-to-end network, named Accelerated DETR, which significantly reduces computational costs while maintaining the advantages of the Transformer architecture, leading to improved detection accuracy. Experimental results demonstrate that Accelerated DETR achieves an impressive AP <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50</inf> of 55.8 and operates at 40 FPS with only 41.6M network parameters.