MIDA-Lab Model Zoo

Welcome to the MIDA-Lab Model Zoo, the central public repository hosting pre-trained deep learning weights for the AutoIS interactive segmentation platforms (Base, Advanced Active Learning, and TRISO Nuclear Fuel Microscopy Suite).

---

📦 Available Model Weights

All official model weights are hosted on the v1.0.0 Release Page.

Model File	Description	Size	File Structure
ucfr_sbd_base.pth	Base Interactive Segmentation Model (SBD Trained)	~378 MB	Single .pth asset
coco_lvis_icl_vit_huge.pth	ViT-Huge Backbone Interactive Segmentation Model	~2.5 GB	Split Assets (.part_aa & .part_ab)

---

🚀 Quick Start: Download & Setup

1. Download Release Assets

You can download assets manually from the v1.0.0 Release Page or use wget in your terminal:

# Navigate to your platform's weights directory
cd src/media/weights/

# Download UCFR Base Model
wget https://github.com/MIDA-Lab/model-zoo/releases/download/v1.0.0/ucfr_sbd_base.pth

# Download ViT-Huge Model Parts
wget https://github.com/MIDA-Lab/model-zoo/releases/download/v1.0.0/coco_lvis_icl_vit_huge.pth.part_aa
wget https://github.com/MIDA-Lab/model-zoo/releases/download/v1.0.0/coco_lvis_icl_vit_huge.pth.part_ab

2. Reassemble Split Weights

Because coco_lvis_icl_vit_huge.pth exceeds standard GitHub asset size thresholds, it is distributed in two parts (.part_aa and .part_ab). Combine them into the complete PyTorch checkpoint:

On Linux / macOS:

cd src/media/weights
cat coco_lvis_icl_vit_huge.pth.part_* > coco_lvis_icl_vit_huge.pth
rm -f coco_lvis_icl_vit_huge.pth.part_*

On Windows:

Warning

Do not use cat ... > ... in Windows PowerShell! PowerShell's > redirection re-encodes binary files as UTF-16 text with a Byte Order Mark (\xff\xfe), which corrupts the checkpoint and causes _pickle.UnpicklingError: invalid load key, '\xff'. Always use binary concatenation:

• Command Prompt (cmd.exe):

  cd src\media\weights
  copy /b coco_lvis_icl_vit_huge.pth.part_aa + coco_lvis_icl_vit_huge.pth.part_ab coco_lvis_icl_vit_huge.pth
  del coco_lvis_icl_vit_huge.pth.part_*

• PowerShell:

  cd src\media\weights
  cmd.exe /c "copy /b coco_lvis_icl_vit_huge.pth.part_aa + coco_lvis_icl_vit_huge.pth.part_ab coco_lvis_icl_vit_huge.pth"
  Remove-Item coco_lvis_icl_vit_huge.pth.part_*

• Git Bash on Windows:

  cd src/media/weights
  cat coco_lvis_icl_vit_huge.pth.part_aa coco_lvis_icl_vit_huge.pth.part_ab > coco_lvis_icl_vit_huge.pth
  rm -f coco_lvis_icl_vit_huge.pth.part_*

---

📄 Citation & Attribution

If you use these pre-trained model weights or algorithms in your research, please cite our publications:

1. CFR-ICL

CFR-ICL: Cascade-Forward Refinement with Iterative Click Loss for Interactive Image Segmentation
arXiv:2303.05620 [cs.CV]
🔗 Paper Link: https://arxiv.org/abs/2303.05620

@article{cfr_icl2023,
  title={CFR-ICL: Cascade-Forward Refinement with Iterative Click Loss for Interactive Image Segmentation},
  author={Shoukun Sun and others},
  journal={arXiv arXiv:2303.05620},
  year={2023},
  url={https://arxiv.org/abs/2303.05620}
}

2. U-CFR

U-CFR: Uncertainty-Guided Cascade Forward Refinement for Interactive Segmentation
Pattern Recognition (Lecture Notes in Computer Science, Springer Nature Switzerland, 2027)
🔗 DOI Link: https://doi.org/10.1007/978-3-032-31666-0_42 | 🔗 arXiv Link: https://arxiv.org/abs/2607.20705

If you find U-CFR useful in your research, please cite our paper:

@InProceedings{10.1007/978-3-032-31666-0_42,
  author    = {Danquah Darko, Elijah and Xian, Min and Soule, Terence and Yao, Tiankai and William Anderson, Matthew},
  editor    = {De Marsico, Maria and Ho, Tin Kam and Jurie, Frederic and Liu, Cheng-Lin and Lopresti, Daniel and Nystr{\"o}m, Ingela and Ogier, Jean-Marc and Ross, Arun and Wang, Liang},
  title     = {U-CFR: Uncertainty-Guided Cascade Forward Refinement for Interactive Segmentation},
  booktitle = {Pattern Recognition},
  series    = {Lecture Notes in Computer Science},
  year      = {2027},
  publisher = {Springer Nature Switzerland},
  address   = {Cham},
  pages     = {645--660},
  doi       = {10.1007/978-3-032-31666-0_42},
  url       = {https://doi.org/10.1007/978-3-032-31666-0_42},
  isbn      = {978-3-032-31666-0}
}

Or cite the arXiv preprint:

@article{danquahdarko2026ucfr,
  title     = {U-CFR: Uncertainty-Guided Cascade Forward Refinement for Interactive Segmentation},
  author    = {Danquah Darko, Elijah and Xian, Min and Soule, Terence and Yao, Tiankai and William Anderson, Matthew},
  journal   = {arXiv preprint arXiv:2607.20705},
  year      = {2026},
  url       = {https://arxiv.org/abs/2607.20705}
}