File size: 3,759 Bytes
3dfa389 f1a7d6b 3dfa389 | 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 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 | ---
license: mit
tags:
- medical-imaging
- mri
- image-translation
- cut
- cyclegan
- stargan-v2
- mrixfields
- baseline
pipeline_tag: image-to-image
library_name: pytorch
---
# MRIxFields2026 Baseline Checkpoints
Pre-trained baseline checkpoints for the [MRIxFields 2026](https://mrixfields.chihucloud.com/2026/) challenge (MICCAI 2026): cross-field brain MRI translation.
## Tasks
| Task | Goal | Input | Output |
|------|------|-------|--------|
| **Task 1** | Ultra-high field synthesis | 0.1T / 1.5T / 3T / 5T | 7T |
| **Task 2** | Low-field enhancement | 0.1T | 1.5T / 3T / 5T / 7T |
| **Task 3** | Any-to-any translation (single unified model) | Any field | Any field |
## Methods
| Method | Paper | Tasks | # sub-tasks |
|--------|-------|-------|-------------|
| **CUT** | Park et al., ECCV 2020 | 1, 2 | 24 |
| **CycleGAN** | Zhu et al., ICCV 2017 | 1, 2 | 24 |
| **StarGAN v2** | Choi et al., CVPR 2020 | 3 | 1 |
Total: **49** checkpoints (one per sub-task × method), **~7 GB**.
## Training
All checkpoints are from the recommended `pro_pretrained` setting: **unpaired pretrain on the retrospective split + paired fine-tune on the prospective split**. This is the strongest baseline; ablation modes (`pro_scratch`, `retro_scratch`) are kept for internal use and are not released.
## Directory Layout
```
.
├── README.md
├── LICENSE
├── task1_<src>_to_7T_<modality>/
│ ├── cut/pro_pretrained/weights/checkpoint_epoch100.pth
│ └── cyclegan/pro_pretrained/weights/checkpoint_epoch100.pth
├── task2_0.1T_to_<tgt>_<modality>/
│ └── (same as task1)
└── task3_any_to_any_multimodal/
└── stargan_v2/pro_pretrained/weights/checkpoint_epoch50.pth
```
> Training-time configs (`config_actual.yaml`) are intentionally not included. Use the config templates in the upstream Baseline repo and override the data paths for your own environment.
## Download
```python
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="mrixfields/MRIxFields2026-Baseline",
repo_type="model",
local_dir="./MRIxFields2026-Baseline",
)
```
Single checkpoint:
```python
from huggingface_hub import hf_hub_download
hf_hub_download(
repo_id="mrixfields/MRIxFields2026-Baseline",
filename="task1_0.1T_to_7T_T1W/cut/pro_pretrained/weights/checkpoint_epoch100.pth",
repo_type="model",
)
```
## Training & Inference
See the upstream Baseline code: [`MRIxFields2026/Baseline`](https://github.com/MRIxFields/MRIxFields2026/tree/main/Baseline).
## License
MIT — see [LICENSE](LICENSE). Copyright © 2026 MRIxFields.
## Citation
If you use these baselines, please cite the original method papers:
```bibtex
@inproceedings{park2020cut,
title={Contrastive Learning for Unpaired Image-to-Image Translation},
author={Taesung Park and Alexei A. Efros and Richard Zhang and Jun-Yan Zhu},
booktitle={European Conference on Computer Vision (ECCV)},
year={2020}
}
@inproceedings{CycleGAN2017,
title={Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks},
author={Zhu, Jun-Yan and Park, Taesung and Isola, Phillip and Efros, Alexei A},
booktitle={IEEE International Conference on Computer Vision (ICCV)},
year={2017}
}
@inproceedings{choi2020starganv2,
title={StarGAN v2: Diverse Image Synthesis for Multiple Domains},
author={Yunjey Choi and Youngjung Uh and Jaejun Yoo and Jung-Woo Ha},
booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2020}
}
```
## Links
- Challenge website: https://mrixfields.chihucloud.com/2026/
- Dataset (Synapse): https://www.synapse.org/Synapse:syn72060672/datasets/
- GitHub: https://github.com/MRIxFields/MRIxFields2026
|