Anchor-Free Oriented Proposal Generator for Object Detection

Oriented object detection is a practical and challenging task in remote sensing image interpretation. Nowadays, oriented detectors mostly use horizontal boxes as intermedium to derive from them. However, the are inclined get small Intersection-over-Unions (IoUs) with ground truths, which may have some undesirable effects, such introducing redundant noise, mismatching detracting robustness of detectors, etc. In this paper, we propose novel Anchor-free Proposal Generator (AOPG) that abandons box-related operations network architecture. AOPG first produces coarse by Coarse Location Module (CLM) an anchor-free manner then refines them into high-quality proposals. After AOPG, apply Fast R-CNN head produce final results. Furthermore, shortage large-scale datasets also hindrance development detection. To alleviate data insufficiency, release new dataset on basis our DIOR name it DIOR-R. Massive experiments demonstrate effectiveness AOPG. Particularly, without bells whistles, achieve accuracy 64.41%, 75.24% 96.22% mAP DIOR-R, DOTA HRSC2016 respectively. Code models available at https://github.com/jbwang1997/AOPG.