hakuuux/VesselSAM_0.2

VesselSAM

VesselSAM adapts SAM with AtrousLoRA for box-prompted aortic vessel segmentation in CT angiography.

• AVT6.pt: checkpoint trained on the Aortic Vessel Tree (AVT) dataset.
• TBAD10.pt: checkpoint trained on the Type-B Aortic Dissection (TBAD) dataset.

These inference-only checkpoints contain adaptation and mask-decoder weights. Both require the same MedSAM ViT-B base checkpoint, downloaded separately. Pass the selected .pt file to --checkpoint and the base weights to --base-checkpoint.

🚀 Inference instructions: GitHub

📄 Also see: Zhao, X., Zhuang, P., Zhang, Z. et al. VesselSAM: a parameter-efficient adaptation of SAM to aortic vessel segmentation with AtrousLoRA. BMC Medical Imaging (2026). DOI.