Fast arbitrary-oriented object detection for remote sensing images

In the filed of remote sensing, arbitrary-oriented object detection methods has gained great attention, benefiting from the accurate detection ability of dense objects. However, the existing methods, which are designed based on ResNet, are not fast enough for real-time application. To solve this problem, our paper proposes a new fast arbitrary-oriented object detection methods based on YOLOX. First, a new head for rotational box prediction is proposed, in which a new branch is designed to extract the angle information through weighted averaging from different angles. Then, a new loss function with sine function is designed to avoid the boundary problem for rotational box prediction. The advantage of this loss is that the value of loss is also periodic which corresponds exactly to the periodicity of rotational box. Experiment results verify that the detection speed of the proposed method is fastest in comparison with the state-of-the-art methods, while maintaining the competitive detection accuracy. Code is available at https://github.com/ljx43031/Fast-AOOD.