Semantic-decoupled Spatial Partition Guided Point-supervised Oriented Object Detection
🔥 We appreciate the attention to our paper. The code is available at Github repo.
Production from Institute of Computing Technology, Chinese Academy of Sciences.
Primary contact: Xinyuan Liu ( [email protected] ).
TL;DR
This repository contains the source code of Semantic-decoupled Spatial Partition Guided Point-supervised Oriented Object Detection.
To tackle inadequate sample assignment and instance confusion in point-supervised oriented object detection for remote sensing dense scenes, we propose SSP (Semantic-decoupled Spatial Partition), a framework integrating rule-driven prior injection and data-driven label purification. Its core innovations include pixel-level spatial partition for sample assignment and semantic-modulated box extraction for pseudo-label generation.
Pseudo-label performance
All pseudo-labeling results are available in pseudo_labels.
| Dataset | mAP | mIoU | ann_file |
|---|---|---|---|
| DOTA-v1.0 | 34.95 | 49.03 | pseudo_labels/ssp_dotav10_hybrid/ |
| DOTA-v1.5 | 28.89 | 44.92 | pseudo_labels/ssp_dotav15_hybrid/ |
| DOTA-v2.0 | 24.72 | 41.93 | pseudo_labels/ssp_dotav20_hybrid/ |
Detectors performance
| Dataset | Config | Log | Checkpoint | mAP(paper) | mAP(reproduced) |
|---|---|---|---|---|---|
| SSP(RFOCS) | config | hugging face | hugging face | 45.78 | 45.82 |
| SSP(ORCNN) | config | hugging face | hugging face | 47.86 | 48.81 |
| SSP(ReDet) | config | hugging face | hugging face | 48.50 | 49.02 |
🖊️ Citation
If you find this work helpful for your research, please consider giving this repo a star ⭐ and citing our papers:
Citation
If this work is helpful for your research, please consider citing the following BibTeX entry.
@misc{liu2025ssp,
title={Semantic-decoupled Spatial Partition Guided Point-supervised Oriented Object Detection},
author={Xinyuan Liu and Hang Xu and Yike Ma and Yucheng Zhang and Feng Dai},
year={2025},
eprint={2506.10601},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2506.10601},
}
@inproceedings{xu2024acm,
title={Rethinking boundary discontinuity problem for oriented object detection},
author={Xu, Hang and Liu, Xinyuan and Xu, Haonan and Ma, Yike and Zhu, Zunjie and Yan, Chenggang and Dai, Feng},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={17406--17415},
year={2024}
}
Related resources
We acknowledge all the open-source contributors for the following projects to make this work possible: