PolarDet: A Fast, More Precise Detector for Rotated Target in Aerial Images

TL;DR

PolarDet is a novel, fast, and precise one-stage detector for rotated objects in aerial images, utilizing a polar coordinate representation to improve accuracy and speed over existing methods.

Contribution

The paper introduces PolarDet, a new detector using polar coordinate representation and a sub-pixel center structure, achieving state-of-the-art accuracy and speed in aerial object detection.

Findings

Achieves state-of-the-art mAP on DOTA, UCAS-AOD, HRSC datasets.

Reaches 32 fps inference speed on UCAS-AOD.

Outperforms previous methods in both accuracy and efficiency.

Abstract

Fast and precise object detection for high-resolution aerial images has been a challenging task over the years. Due to the sharp variations on object scale, rotation, and aspect ratio, most existing methods are inefficient and imprecise. In this paper, we represent the oriented objects by polar method in polar coordinate and propose PolarDet, a fast and accurate one-stage object detector based on that representation. Our detector introduces a sub-pixel center semantic structure to further improve classifying veracity. PolarDet achieves nearly all SOTA performance in aerial object detection tasks with faster inference speed. In detail, our approach obtains the SOTA results on DOTA, UCAS-AOD, HRSC with 76.64\% mAP, 97.01\% mAP, and 90.46\% mAP respectively. Most noticeably, our PolarDet gets the best performance and reaches the fastest speed(32fps) at the UCAS-AOD dataset.