CephVIT

CephVIT is a deep learning application for automatic cephalometric landmark detection from cephalogram images.

Requirements

CephViT requires the following Python packages:

torch
torchvision
torchaudio
timm
gradio
opencv-python
numpy
pillow
albumentations
cryptography
huggingface_hub
pandas
pydantic
tqdm

All dependencies are included in requirements.txt and can be installed with:

pip install -r requirements.txt

Run Locally

Clone the repository:

git clone https://huggingface.co/nlm-dir/CephViT
cd CephViT

Install dependencies:

pip install -r requirements.txt

The included model checkpoint is encrypted. Set the MODEL_KEY environment variable before running:

export MODEL_KEY="2d65697cf6ad50c95a0a2558c73658a7"

Then start the app:

python app.py

Project Structure

CephVIT/
├── app.py                  # Gradio inference application
├── model.py                # Model definitions
├── heatmap_utils.py        # Heatmap generation and decoding
├── secure_torch_load.py    # Encrypted checkpoint loader
├── best.pt.enc             # Encrypted model checkpoint
└── requirements.txt

License

OpenMDW 1.1.

Citation

If you use CephViT in your work, please cite:

@inproceedings{hou2026automatic,
  title     = {Automatic Cephalometric Landmark Localization on {CBCT}-Derived Digitally Reconstructed Radiographs for Skeletal Malocclusion Classification},
  author    = {Hou, Benjamin and Almpani, Konstantinia and Lee, Janice S. and Lu, Zhiyong},
  booktitle = {Oral and Dental Image Analysis (ODIN 2026)},
  series    = {Lecture Notes in Computer Science},
  publisher = {Springer},
  year      = {2026},
  note      = {Accepted for publication}
}

Accepted to ODIN 2026. Citation details will be updated when the Springer proceedings are published.