DRDet: Dual-Angle Rotated Line Representation for Oriented Object Detection

In aerial scenes, oriented object detection is sensitive to the orientation of objects, which makes the formulation of orientation-aware object representation become a critical problem. Existing methods mostly adopt rectangle anchors or discrete points as object representation, which may lead to the feature aliasing between overlapping objects and ignore the orientation information of objects. To solve these issues, we propose a novel anchor-free oriented object detection network named DRDet, which adopts dual-angle rotated lines (DRLs) as object representation. Different from other object representations, DRL can adaptively rotate and extend to the boundary of the object according to its orientation and shape, which explicitly introduces the orientation information into the formulation of object representation. And it can adaptively cope with the geometric deformation of objects. Based on the DRLs, we design an orientation-guided feature encoder (OFE) to encode discriminant object features along each rotated line, respectively. Instead of encoding the rectangle feature, the OFE module adopts line features for orientation-guided feature encoding, which can alleviate the feature aliasing between neighboring objects or backgrounds. To further enhance the flexibility of DRLs, we design a dual-angle decoder (DD) that predicts two angle offsets according to the orientation-guided feature and converts the angle offsets and regression offsets into DRL representation, which can help to guide the adaptive rotation of each rotated line, respectively. Our proposed method achieves consistent improvement on both DOTA and HRSC2016 datasets. Extensive experiment results verify the effectiveness of our method in oriented object detection.

Automated summary

In this paper, a novel anchor-free oriented object detection network called DRDet is proposed, which adopts dual-angle rotated lines (DRLs) as object representation. By introducing DRL, the network can adaptively rotate and extend to the boundary of the object, explicitly incorporating orientation information into the formulation of object representation. Furthermore, an orientation-guided feature encoder (OFE) and a dual-angle decoder (DD) are designed to enhance the flexibility and performance of DRLs. Experimental results demonstrate consistent improvement in oriented object detection using the proposed method.