SegmentAnyTooth

An open-source deep learning framework for tooth enumeration and segmentation in intraoral photos.

---

📢 Update (2025-01-17)

We are pleased to announce that our paper has been accepted in the Journal of Dental Sciences.

Citation:

Nguyen, K. D., Hoang, H. T., Doan, T.-P. H., Dao, K. Q., Wang, D.-H., & Hsu, M.-L. (2025).
SegmentAnyTooth: An open-source deep learning framework for tooth enumeration and segmentation in intraoral photos.
Journal of Dental Sciences, S1991790225000030. https://doi.org/10.1016/j.jds.2025.01.003

---

🚀 Installation

We use uv as the dependency manager.

All dependencies are specified in pyproject.toml.
Please prefer using pyproject.toml directly to manage the environment.

Example installation:

pip install uv
uv pip install -r pyproject.toml

(Optionally) Install development dependencies:

uv pip install -r pyproject.toml --group dev

Weights download

The weight is released under SegmentAnyTooth Non-Commercial License.

To obtain the weight, please sign the agreement form and email the signed version to [email protected]. I will email the link for weight download in working days.

🚀 Usage Example

After installing the dependencies, you can run predictions:

from segmentanytooth import predict

# Predict tooth masks from an intraoral image
mask = predict(
    image_path="path/to/your/image.png",
    view="upper",            # one of: "upper", "lower", "left", "right", "front"
    weight_dir="./weights",  # path to downloaded model weights
    sam_batch_size=10,       # optional: adjust batch size for faster SAM inference
)

# Save the predicted mask
import cv2
cv2.imwrite("predicted_mask.jpg", mask * 5)  # Scale mask for visualization if needed

Notes:

• Model weights must be placed in the weights/ directory (or specify another directory).
• Views correspond to different photographic angles of intraoral images.
• Output mask is a NumPy array with pixel values representing FDI tooth numbers.

📜 License

• Code is licensed under the MIT License. You are free to use, modify, and distribute the code with attribution.
• Pretrained model weights are provided under a Non-Commercial Use License. Commercial use of the weights is prohibited without explicit permission.

Please see:

• LICENSE for full terms of the MIT License.
• WEIGHT_LICENSE for the model weights licensing terms.

📚 Citation

If you use SegmentAnyTooth in your work, please cite:

@article{Nguyen2025SegmentAnyTooth,
  title={SegmentAnyTooth: An open-source deep learning framework for tooth enumeration and segmentation in intraoral photos},
  author={Nguyen, Khoa D. and Hoang, Huy T. and Doan, Thi-Phuong-Hoa and Dao, Kim-Quyen and Wang, Ding-Han and Hsu, Min-Ling},
  journal={Journal of Dental Sciences},
  year={2025},
  doi={10.1016/j.jds.2025.01.003}
}