This repository implements R3Det and R3Det++, deep learning-based single-stage object detectors published at AAAI 2021. The system uses a Feature Refinement Module (FRM) to iteratively refine detection on oriented objects, supporting datasets like DOTA and HRSC2016. Built on RetinaNet with Focal Loss, it includes techniques such as IoU-Smooth L1 Loss and Circular Smooth Label (CSL) for rotation-aware detection.