MonaiLabel

Download Models

Initialize MedSAM Bundle's model from huggingface:

docker compose exec -it monailabel bash
python # start a python interpreter in monailabel container

from transformers import SamModel, SamProcessor
import torch

MonaiLabel requires model.pt and model_best.pt in the bundle:

model = SamModel.from_pretrained("flaviagiammarino/medsam-vit-base", local_files_only=False)
torch.save(model.state_dict(), '/monailabel/apps/monaibundle/model/MedSamBundle/models/model.pt')
torch.save(model.state_dict(), '/monailabel/apps/monaibundle/model/MedSamBundle/models/model_best.pt')

Loading weights if needed for custom models:

# loaded_weights = torch.load('/monailabel/apps/monaibundle/model/MedSamBundle/models/model.pt', weights_only=True)
# model.load_state_dict(loaded_weights)

Restart MonaiLabel

docker compose restart monailabel