File size: 3,942 Bytes
90543bf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
---
license: apache-2.0
---
# Semantic-decoupled Spatial Partition Guided Point-supervised Oriented Object Detection

[![arxiv](https://img.shields.io/badge/arXiv-2506.10601-479ee2.svg)](https://arxiv.org/pdf/2506.10601)
[![Github](https://img.shields.io/badge/GitHub-SSP-blueviolet.svg)](https://github.com/antxinyuan/ssp)

🔥 We appreciate the attention to our paper. The code is available at [Github repo](https://github.com/antxinyuan/ssp).

> Production from Institute of Computing Technology, Chinese Academy of Sciences.

> Primary contact: Xinyuan Liu ( liuxinyuan21s@ict.ac.cn ).

## TL;DR
This repository contains the source code of [**Semantic-decoupled Spatial Partition Guided Point-supervised Oriented Object Detection**](https://arxiv.org/pdf/2506.10601).

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](https://huggingface.co/antxinyuan/SSP/blob/main/pseudo_labels.zip).

| 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](https://github.com/antxinyuan/ssp/blob/main/configs/ssp/rfcos_ssp_dotav10.py) | [hugging face](https://huggingface.co/antxinyuan/SSP/blob/main/logs/rfcos_ssp_dotav10.json) |  [hugging face](https://huggingface.co/antxinyuan/SSP/blob/main/models/rfcos_ssp_dotav10-4c17ff33.pth) | 45.78 | 45.82 |
| SSP(ORCNN) | [config](https://github.com/antxinyuan/ssp/blob/main/configs/ssp/orcnn_ssp_dotav10.py) |  [hugging face](https://huggingface.co/antxinyuan/SSP/blob/main/logs/orcnn_ssp_dotav10.json) | [hugging face](https://huggingface.co/antxinyuan/SSP/blob/main/models/orcnn_ssp_dotav10-2df034d3.pth) | 47.86 | 48.81 |
| SSP(ReDet) | [config](https://github.com/antxinyuan/ssp/blob/main/configs/ssp/orcnn_ssp_dotav20.py) |  [hugging face](https://huggingface.co/antxinyuan/SSP/blob/main/logs/redet_ssp_dotav10.json) | [hugging face](https://huggingface.co/antxinyuan/SSP/blob/main/models/redet_ssp_dotav10-eed2738e.pth) | 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.

``` bibtex
@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:
- [PointOBB-v2](https://github.com/VisionXLab/PointOBB-v2)
- [MMRotate](https://github.com/open-mmlab/mmrotate)