Installation

1. Create a virtual environment conda create -n medsam python=3.10 -y and activate it conda activate medsam
2. Install pip3 install torch torchvision torchaudio
3. git clone https://github.com/DLYuanGod/BioSAM2
4. Enter the BioSAM2 folder cd BioSAM2 and run pip install -e .
5. Enter cd segment-anything-2 and pip install -e . ; cd ..

Model Training

Data preprocessing

Download [SAM2 checkpoint] cd segment-anything-2/checkpoints;./download_ckpts.sh and place it at here and change the cfg here.

Download the dataset.

Run pre-processing

Datasets 701 and 702:

Install cc3d: pip install connected-components-3d and change the path here and here and output dir here.

python pre_CT_MR.py

Datasets 703 and 704

Change the path here and here and output dir here.

python pre_CELL_EN.py

Training on one GPU

Change the data npy path here and here.

python train_one_gpu.py

Testing

Change the ckpt here and arg here.

And eval using