Learning Point-Guided Localization for Detection in Remote Sensing Images

Object detection in remote sensing images is challenging due to the dense distribution and arbitrary angle of objects. It a consensus that oriented bounding box (OBB) more suitable represent aerial However, there are some extreme cases regression-based OBB make regression target discontinuous, resulting poor performance. In this article, an analysis formats problems its presented, following with exploration transform localization from keypoint estimation, which could be applied avoid problem discontinuous target. Our novel method called Object-wise Point-guided Localization Detector (OPLD). Continuously, new prediction center-point introduced refine results, as truncation caused by cut graph. Lastly, order figure inconsistency between quality classification score, both endpoint scores score adopted weighting result score. Experimental results based on two widely used datasets, i.e., DOTA HRSC2016. OPLD achieve 76.43% mAP 78.35% horizontal boxes tasks DOTA-v1.0, achieves state-of-the-art performance, respectively. Project page at https://github.com/yf19970118/OPLD-Pytorch.