|
|
--- |
|
|
license: cc-by-4.0 |
|
|
datasets: |
|
|
- AnonRes/OpenMind |
|
|
pipeline_tag: image-feature-extraction |
|
|
tags: |
|
|
- medical |
|
|
--- |
|
|
|
|
|
# OpenMind Benchmark 3D SSL Models |
|
|
|
|
|
> **Model from the paper**: [An OpenMind for 3D medical vision self-supervised learning](https://arxiv.org/abs/2412.17041) |
|
|
> **Pre-training codebase used to create checkpoint**: [MIC-DKFZ/nnssl](https://github.com/MIC-DKFZ/nnssl) |
|
|
> **Dataset**: [AnonRes/OpenMind](https://huggingface.co/datasets/AnonRes/OpenMind) |
|
|
> **Downstream (segmentation) fine-tuning**: [TaWald/nnUNet](https://github.com/TaWald/nnUNet) |
|
|
|
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
## 🔍 Overview |
|
|
|
|
|
This repository hosts pre-trained checkpoints from the **OpenMind** benchmark: |
|
|
📄 **"An OpenMind for 3D medical vision self-supervised learning"** |
|
|
([arXiv:2412.17041](https://arxiv.org/abs/2412.17041)) — the first extensive benchmark study for **self-supervised learning (SSL)** on **3D medical imaging** data. |
|
|
|
|
|
The models were pre-trained using various SSL methods on the [OpenMind Dataset](https://huggingface.co/datasets/AnonRes/OpenMind), a large-scale, standardized collection of public brain MRI datasets. |
|
|
|
|
|
**These models are not recommended to be used as-is.** Instead we recommend using the downstream fine-tuning pipelines for **segmentation** and **classification**, available in the [adaptation repository](https://github.com/TaWald/nnUNet). |
|
|
*While direct download is possible, we recommend using the auto-download of the respective fine-tuning repositories.* |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧠 Model Variants |
|
|
|
|
|
We release SSL checkpoints for two backbone architectures: |
|
|
|
|
|
- **ResEnc-L**: A CNN-based encoder [[link1](https://arxiv.org/abs/2410.23132), [link2](https://arxiv.org/abs/2404.09556)] |
|
|
- **Primus-M**: A transformer-based encoder [[Primus paper](https://arxiv.org/abs/2503.01835)] |
|
|
|
|
|
Each encoder has been pre-trained using the following SSL techniques: |
|
|
|
|
|
| Method | Description | |
|
|
|---------------|-------------| |
|
|
| [Volume Contrastive (VoCo)](https://arxiv.org/abs/2402.17300) | Global contrastive learning in 3D volumes | |
|
|
| [VolumeFusion (VF)](https://arxiv.org/abs/2306.16925) | Spatial fusion-based SSL | |
|
|
| [Models Genesis (MG)](https://www.sciencedirect.com/science/article/pii/S1361841520302048) | Classic 3D self-reconstruction | |
|
|
| [Masked Autoencoders (MAE)](https://openaccess.thecvf.com/content/CVPR2022/html/He_Masked_Autoencoders_Are_Scalable_Vision_Learners_CVPR_2022_paper) | Patch masking and reconstruction | |
|
|
| [Spark 3D (S3D)](https://arxiv.org/abs/2410.23132) | 3D adaptation of Spark framework | |
|
|
| [SimMIM](https://openaccess.thecvf.com/content/CVPR2022/html/Xie_SimMIM_A_Simple_Framework_for_Masked_Image_Modeling_CVPR_2022_paper.html) | Simple masked reconstruction | |
|
|
| [SwinUNETR SSL](https://arxiv.org/abs/2111.14791) | Transformer-based pre-training | |
|
|
| [SimCLR](https://arxiv.org/abs/2002.05709) | Contrastive learning baseline | |
|
|
|
