[Feature] Support Semi-supervised Oriented Object Detection: SOOD (CVPR 2023)

[Haru-zt]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Support Semi-supervised Oriented Object Detection: SOOD (CVPR 2023).
SOOD is based on mmrotate 0.x, but not work in mmrotate 1.x

Modification

1. SOOD configs files (including data augmentation)
2. Semi-base
3. SOOD code
4. OT loss used in SOOD
5. Mean Teacher Hook supports Burn-in strategy
6. provide README.md to split train set and start training

BC-breaking (Optional)

No

Use cases (Optional)

please refer to sood/README.md

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues. Yes
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. The documentation has been modified accordingly, like docstring or example tutorials. Yes

[CLAassistant]

All committers have signed the CLA.

[yangxue0827]

The source code seems to be based on MMRotate 0.x. Can 1.x reproduce the performance in the article? It seems that the README.md lacks relevant information. We also welcome SOOD to be merged into the main (0.x) branch.

[yangxue0827]

mAP？

[Haru-zt]

hi I only reproduce the results on 10% setting and i got 47.93 while the article report 48.63. I think that is a fair result. As I am not the author of the article and more important i am not familiar with the 0.x, it is hard to merge on old framework. 

…

---Original--- From: ***@***.***> Date: Thu, Mar 20, 2025 21:08 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [open-mmlab/mmrotate] [Feature] Support Semi-supervised OrientedObject Detection: SOOD (CVPR 2023) (PR #1003) @yangxue0827 commented on this pull request. The source code seems to be based on MMRotate 0.x. Can 1.x reproduce the performance in the article? It seems that the README.md lacks relevant information. We also welcome SOOD to be merged into the main (0.x) branch. In configs/sood/README.md: > +│ │ │ ├── annfiles +│ │ ├── train_30_labeled +│ │ │ ├── images +│ │ │ ├── annfiles +│ │ ├── train_30_unlabeled +│ │ │ ├── images +│ │ │ ├── annfiles +│ │ ├── val +│ │ │ ├── images +│ │ │ ├── annfiles +│ │ ├── test +│ │ │ ├── images +│ │ │ ├── annfiles +``` + +## Results mAP？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[yangxue0827]

hi I only reproduce the results on 10% setting and i got 47.93 while the article report 48.63. I think that is a fair result. As I am not the author of the article and more important i am not familiar with the 0.x, it is hard to merge on old framework. 
…
---Original--- From: @.> Date: Thu, Mar 20, 2025 21:08 PM To: @.>; Cc: @.@.>; Subject: Re: [open-mmlab/mmrotate] [Feature] Support Semi-supervised OrientedObject Detection: SOOD (CVPR 2023) (PR #1003) @yangxue0827 commented on this pull request. The source code seems to be based on MMRotate 0.x. Can 1.x reproduce the performance in the article? It seems that the README.md lacks relevant information. We also welcome SOOD to be merged into the main (0.x) branch. In configs/sood/README.md: > +│ │ │ ├── annfiles +│ │ ├── train_30_labeled +│ │ │ ├── images +│ │ │ ├── annfiles +│ │ ├── train_30_unlabeled +│ │ │ ├── images +│ │ │ ├── annfiles +│ │ ├── val +│ │ │ ├── images +│ │ │ ├── annfiles +│ │ ├── test +│ │ │ ├── images +│ │ │ ├── annfiles +``` + +## Results mAP？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

Could you update the README with the 10% results?

[Haru-zt]

Fine!

[yangxue0827]

LGTM