🧠 Depthwise-Dilated Convolutional Adapters for Medical Object Tracking and Segmentation Using the Segment Anything Model 2 (DD-SAM2)

---

📁 Repository Structure

Please refer to the following key scripts and modules for implementation details:

• sam2/adapter_ap.py — contains the core DD_Adapter implementation.
• train_xx.py — training script for DD-SAM2.
• test_xx.py — evaluation and inference script.
• save_seg_result_xx.py — script for saving segmentation and tracking results.

---

🧩 About DD-SAM2

Our framework builds upon SAM2 and MedSAM2, integrating depthwise-dilated convolutional adapters to enhance feature representation for medical object tracking and segmentation tasks.

---

📚 Citation

If you find this work helpful, please cite our paper along with SAM2 and MedSAM2.

@article{xu2025depthwise,
  title   = {Depthwise-dilated convolutional adapters for medical object tracking and segmentation using the Segment Anything Model 2},
  author  = {Xu, Guoping and Kabat, Christopher and Zhang, You},
  journal = {Machine Learning: Science and Technology},
  year    = {2025}
}

Paper Links:

• arXiv Preprint
• MLST Journal Version

---

⚖️ Acknowledgements

Our implementation is based on:

• Segment Anything Model 2 (SAM2)
• MedSAM2

Please cite these works if you use DD-SAM2 in your research.