Abstract:Detecting objects in remote sensing images (RSIs) using oriented bounding boxes (OBBs) is flourishing but challenging, wherein the design of OBB representations is the key to achieving accurate detection. In this article, we focus on two issues that hinder the performance of the two-stage oriented detectors: 1) the notorious boundary discontinuity problem, which would result in significant loss increases in boundary conditions, and 2) the inconsistency in regression schemes between the two stages. We propose a simple and effective bounding box representation by drawing inspiration from the polar coordinate system and integrate it into two detection stages to circumvent the two issues. The first stage specifically initializes four quadrant points as the starting points of the regression for producing high-quality oriented candidates without any postprocessing. In the second stage, the final localization results are refined using the proposed novel bounding box representation, which can fully release the capabilities of the oriented detectors. Such consistency brings a good trade-off between accuracy and speed. With only flipping augmentation and single-scale training and testing, our approach with ResNet-50-FPN harvests 76.25% mAP on the DOTA dataset with a speed of up to 16.5 frames/s, achieving the best accuracy and the fastest speed among the mainstream two-stage oriented detectors. Additional results on the DIOR-R and HRSC2016 datasets also demonstrate the effectiveness and robustness of our method. The source code is publicly available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/yanqingyao1994/QPDet</uri> .

CrossDet++: Growing Crossline Representation for Object Detection

CrossDet - Crossline Representation for Object Detection.

DE-CrossDet: Divisible and Extensible Crossline Representation for Object Detection

A Transformer-Based Object Detector with Coarse-Fine Crossing Representations

DCDet: Dynamic Cross-based 3D Object Detector

DetNet: A Backbone network for Object Detection

DetNet: Design Backbone for Object Detection

Cross-modal Deformable DETR for RGB-D Object Detection

Global Guided Cross-Modal Cross-Scale Network for RGB-D Salient Object Detection

Oriented Object Detection via Contextual Dependence Mining and Penalty-Incentive Allocation

CrossRectify: Leveraging Disagreement for Semi-supervised Object Detection

An Improved DETR Based on Angle Denoising and Oriented Boxes Refinement for Remote Sensing Object Detection

A MultiPath Network for Object Detection

DetOH: An Anchor-Free Object Detector with Only Heatmaps.

Cross-Grid Label Assignment for Arbitrary-Oriented Object Detection in Aerial Images

CSSDet: small object detection via cross-scale feature enhancement on drone-view images

Cross-modal refined adjacent-guided network for RGB-D salient object detection

EfficientDet: Scalable and Efficient Object Detection

SimDet: Cross Similarity Attention for One-shot Object Detection

C $^{2}$ DFNet: Criss-Cross Dynamic Filter Network for RGB-D Salient Object Detection

On Improving Bounding Box Representations for Oriented Object Detection