SSL3D-checkpoints / README.md
consti95's picture
Model card: complete HiLab (Vox-MMSD) citation
6ad2c6c verified
|
Raw History Blame Contribute Delete
9.78 kB
---
license: cc-by-4.0
tags:
- self-supervised-learning
- medical-imaging
- 3d
- nnssl
- ssl3d-challenge
datasets:
- MIC-DKFZ/OpenMind
---
# SSL3D Challenge – Pretrained Checkpoints
Pretrained encoder checkpoints submitted to the **SSL3D challenge** (3D self-supervised learning for medical imaging), plus the organizer MAE baseline.
All models were pretrained on the **OpenMind** dataset.
## Layout
```
<track>/<team>/<team>_<model>.pth
```
- `ResEnc-L`: CNN track (nnU-Net ResEnc-L encoder)
- `Primus-M`: Transformer track (Primus-M encoder)
Files are renamed to `<team>_<model>.pth` (model = `ResEncL` or `PrimusM`); the original submitted file names are listed in the table below.
## Usage
For using the checkpoints, see [MIC-DKFZ/nnssl](https://github.com/MIC-DKFZ/nnssl), which links the segmentation and classification fine-tuning repositories.
Fine-tuning:
- **Segmentation:** [nnU-Net – Fine-tuning from nnssl checkpoints](https://github.com/MIC-DKFZ/nnUNet/blob/master/documentation/finetuning_from_nnssl_checkpoints.md)
- **Classification:** [image_classification – nnFoundation](https://github.com/MIC-DKFZ/image_classification/blob/master/docs/nnfoundation.md)
## Checkpoints
| Track | Team | Method (as in the manuscript) | File | Original submitted file | Code | Paper |
|---|---|---|---|---|---|---|
| ResEnc-L | Datamen | VoxCon | `ResEnc-L/Datamen/Datamen_ResEncL.pth` | `Datamen_resencl_weights_2025-08-23-02-18.pth` | [pwesp/voxcon](https://github.com/pwesp/voxcon) | – |
| ResEnc-L | Organizers (MAE baseline) | MAE | `ResEnc-L/DKFZ_MAE/DKFZ_MAE_ResEncL.pth` | `DKFZ_MAE_DKFZ_BaseMAEResEncL_Openmind_checkpoint_final_2025-09-08-16-33.pth` | [MIC-DKFZ/nnssl](https://github.com/MIC-DKFZ/nnssl) | Wald et al., OpenMind (ICCV) |
| ResEnc-L | HiLab | Voxel-wise Multi-scale Self-Distillation (Vox-MSD) | `ResEnc-L/HiLab/HiLab_ResEncL.pth` | `HiLab_Test_submission_ResEncL_2025-08-14-09-23.pth` | – | Zhou et al. 2025, IEEE J. Biomed. Health Inform. (Vox-MMSD) |
| ResEnc-L | Isikun.MMLab | Variational Contrastive Learning (VCL) | `ResEnc-L/Isikun.MMLab_and_C.H.Aarau.Radiology/Isikun.MMLab_and_C.H.Aarau.Radiology_ResEncL.pth` | `Isikun.MMLab_and_C.H.Aarau.Radiology_vcl_resencl_final_2025-08-20-10-13.pth` | [verimsu/VCL](https://github.com/verimsu/VCL) | Yavuz & Yanikoglu 2024, IEEE FG |
| ResEnc-L | NTUST-EE304 | VoCo + MAE | `ResEnc-L/NTUST_EE304/NTUST_EE304_ResEncL.pth` | `NTUST_EE304_VoCo+MAE_checkpoint_final_2025-08-13-04-38.pth` | [HenryLiu0402/nnssl_VoCo-plus-MAE](https://github.com/HenryLiu0402/nnssl_VoCo-plus-MAE) | – |
| ResEnc-L | Pep3Crew | Anatomically-Grounded Disentanglement | `ResEnc-L/Pep3Crew/Pep3Crew_ResEncL.pth` | `Pep3Crew_checkpoint_best_ExtendedHealth_2208_1620_2025-08-22-20-31.pth` | [jbanusco/BrainFM4Challenges](https://github.com/jbanusco/BrainFM4Challenges/tree/ssl3d_model) (branch `ssl3d_model`) | Gordaliza et al. 2026, [arXiv:2601.13166](https://arxiv.org/abs/2601.13166) |
| ResEnc-L | PVG | BigSimCLR-3D | `ResEnc-L/PVG/PVG_ResEncL.pth` | `PVG_checkpoint_latest_epoch190_2025-08-22-21-50.pth` | [emilykaczmarek/SSL3D_Challenge](https://github.com/emilykaczmarek/SSL3D_Challenge) | – |
| ResEnc-L | Anonymous | Generalized Masked Autoencoder (GMAE) | `ResEnc-L/Anonymous/Anonymous_ResEncL.pth` | – | – | – |
| ResEnc-L | The Latent Campus | Consistent View Alignment (CVA) | `ResEnc-L/The_Latent_Campus/The_Latent_Campus_ResEncL.pth` | `The_Latent_Campus_cnn_cva_contrastive_mae_2025-08-21-08-27.pth` | [Tenbatsu24/LatentCampus](https://github.com/Tenbatsu24/LatentCampus) | Vaish et al. 2025, [arXiv:2509.13846](https://arxiv.org/abs/2509.13846) |
| Primus-M | Datamen | VoxCon | `Primus-M/Datamen/Datamen_PrimusM.pth` | `Datamen_primusm_weights_2025-08-23-08-09.pth` | [pwesp/voxcon](https://github.com/pwesp/voxcon) | – |
| Primus-M | Organizers (MAE baseline) | MAE | `Primus-M/DKFZ_MAE/DKFZ_MAE_PrimusM.pth` | `DKFZ_MAE_DKFZ_BaseMAEPrimusM_Openmind_checkpoint_final_2025-09-08-16-48.pth` | [MIC-DKFZ/nnssl](https://github.com/MIC-DKFZ/nnssl) | Wald et al., OpenMind (ICCV) |
| Primus-M | gamma | DINOv2 Self-Distillation | `Primus-M/gamma/gamma_PrimusM.pth` | `gamma_teacher_74999_2025-08-22-15-27.pth` | [xdoni4/ct-dino](https://github.com/xdoni4/ct-dino) | – |
| Primus-M | HiLab | Voxel-wise Multi-scale Self-Distillation (Vox-MSD) | `Primus-M/HiLab/HiLab_PrimusM.pth` | `HiLab_Test_submission_Primus_2025-08-14-13-14.pth` | – | Zhou et al. 2025, IEEE J. Biomed. Health Inform. (Vox-MMSD) |
| Primus-M | Isikun.MMLab | Variational Contrastive Learning (VCL) | `Primus-M/Isikun.MMLab_and_C.H.Aarau.Radiology/Isikun.MMLab_and_C.H.Aarau.Radiology_PrimusM.pth` | `Isikun.MMLab_and_C.H.Aarau.Radiology_vcl_prismusm_final_2025-08-20-10-18.pth` | [verimsu/VCL](https://github.com/verimsu/VCL) | Yavuz & Yanikoglu 2024, IEEE FG |
| Primus-M | NTUST-EE304 | MAE-filtered | `Primus-M/NTUST_EE304/NTUST_EE304_PrimusM.pth` | `NTUST_EE304_MAE-filtered-1k_checkpoint_final_2025-08-22-07-08.pth` | [HenryLiu0402/nnssl_MAE-filtered-1k](https://github.com/HenryLiu0402/nnssl_MAE-filtered-1k) | – |
| Primus-M | Anonymous | Generalized Masked Autoencoder (GMAE) | `Primus-M/Anonymous/Anonymous_PrimusM.pth` | – | – | – |
| Primus-M | The Latent Campus | Consistent View Alignment (CVA) | `Primus-M/The_Latent_Campus/The_Latent_Campus_PrimusM.pth` | `The_Latent_Campus_eva_cva_contrastive_mae_2025-08-21-08-40.pth` | [Tenbatsu24/LatentCampus](https://github.com/Tenbatsu24/LatentCampus) | Vaish et al. 2025, [arXiv:2509.13846](https://arxiv.org/abs/2509.13846) |
## Pretraining data
OpenMind (large-scale 3D brain MRI dataset; Wald et al., ICCV).
## License
All checkpoints are released under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/), the same license as the [OpenMind dataset](https://huggingface.co/datasets/MIC-DKFZ/OpenMind) they were pretrained on. When using a checkpoint, please cite the works listed under [Citation](#citation).
## Citation
If you use any of these checkpoints, please cite:
1. **OpenMind** (pretraining dataset and nnssl framework): T. Wald, C. Ulrich, J. Suprijadi, S. Ziegler, M. Nohel, R. Peretzke, G. Köhler, K. H. Maier-Hein. *An OpenMind for 3D medical vision self-supervised learning.* ICCV 2025. [arXiv:2412.17041](https://arxiv.org/abs/2412.17041)
2. **SSL3D challenge paper:** *to be added after publication.*
3. **The method paper of the checkpoint you use**, where one exists:
| Team | Method | Citation |
|---|---|---|
| The Latent Campus | CVA | P. Vaish, F. Meister, T. Heimann, C. Brune, J. M. Wolterink. *Consistent View Alignment Improves Foundation Models for 3D Medical Image Segmentation.* 2025. [arXiv:2509.13846](https://arxiv.org/abs/2509.13846) |
| Pep3Crew | Anatomically-Grounded Disentanglement | P. M. Gordaliza, J. Banus, B. Gérin, M. Wynen, N. Molchanova, J. Richiardi, M. Bach Cuadra. *From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models.* 2026. [arXiv:2601.13166](https://arxiv.org/abs/2601.13166) |
| HiLab | Vox-MSD | Y. Zhou, J. Wu, J. Fu, Q. Yue, W. Liao, S. Zhang, S. Zhang, G. Wang. *Vox-MMSD: Voxel-Wise Multi-Scale and Multi-Modal Self-Distillation for Self-Supervised Brain Tumor Segmentation.* IEEE Journal of Biomedical and Health Informatics, 30(9):7966–7979, 2026. [doi:10.1109/JBHI.2025.3592116](https://doi.org/10.1109/JBHI.2025.3592116) |
| Isikun.MMLab | VCL | M. C. Yavuz, B. Yanikoglu. *Self-supervised variational contrastive learning with applications to face understanding.* IEEE FG 2024. [arXiv:2312.00824](https://arxiv.org/abs/2312.00824) |
| Organizers | MAE | see OpenMind (1.) |
| NTUST-EE304, PVG, gamma, Datamen, Anonymous | – | No paper yet; please reference the team's code repository listed in the checkpoint table. |
```bibtex
@inproceedings{wald2025openmind,
title = {An OpenMind for 3D medical vision self-supervised learning},
author = {Wald, Tassilo and Ulrich, Constantin and Suprijadi, Jonathan and Ziegler, Sebastian and Nohel, Michal and Peretzke, Robin and K{\"o}hler, Gregor and Maier-Hein, Klaus H.},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
year = {2025},
eprint = {2412.17041},
archivePrefix = {arXiv}
}
@article{vaish2025cva,
title = {Consistent View Alignment Improves Foundation Models for 3D Medical Image Segmentation},
author = {Vaish, Puru and Meister, Felix and Heimann, Tobias and Brune, Christoph and Wolterink, Jelmer M.},
journal = {arXiv preprint arXiv:2509.13846},
year = {2025}
}
@article{gordaliza2026lessons,
title = {From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models},
author = {Gordaliza, Pedro M. and Banus, Jaume and G{\'e}rin, Beno{\^\i}t and Wynen, Maxence and Molchanova, Nataliia and Richiardi, Jonas and Bach Cuadra, Meritxell},
journal = {arXiv preprint arXiv:2601.13166},
year = {2026}
}
@article{zhou2026voxmmsd,
title = {Vox-MMSD: Voxel-Wise Multi-Scale and Multi-Modal Self-Distillation for Self-Supervised Brain Tumor Segmentation},
author = {Zhou, Yubo and Wu, Jianghao and Fu, Jia and Yue, Qiang and Liao, Wenjun and Zhang, Shichuan and Zhang, Shaoting and Wang, Guotai},
journal = {IEEE Journal of Biomedical and Health Informatics},
volume = {30},
number = {9},
pages = {7966--7979},
year = {2026},
doi = {10.1109/JBHI.2025.3592116}
}
@inproceedings{yavuz2024vcl,
title = {Self-supervised variational contrastive learning with applications to face understanding},
author = {Yavuz, Mehmet Can and Yanikoglu, Berrin},
booktitle = {IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG)},
year = {2024}
}
```