Awesome

Biomedical SAM-2: Segment Anything in Biomedical Images and Videos

Biomedical SAM-2 (BioSAM-2) is an enhanced foundation model optimized for biomedical data based on SAM-2.

Installation

Download code:

git clone https://github.com/ZhilingYan/Biomedical-SAM-2.git
cd Biomedical-SAM-2

Install the environment:

conda env create -f environment.yml
conda activate BioSAM2

Then we need to download a model checkpoint. All the model checkpoints can be downloaded by running:

cd checkpoints && \
./download_ckpts.sh && \
cd ..

or manually dowload Hiera-s from: sam2_hiera_small.pt

Model Training

Downloading BUSI dataset from here

Train and validate the dataset:

python train.py -net sam2 -exp_name BUSI -vis 1 -sam_ckpt ./checkpoints/sam2_hiera_small.pt -sam_config sam2_hiera_s -image_size 1024 -out_size 1024 -b 4 -val_freq 1 -dataset BUSI -data_path BUSI_PATH -DiceCEloss True -nprompt click

Cite

@article{yan2024biomedical,
  title={Biomedical sam 2: Segment anything in biomedical images and videos},
  author={Yan, Zhiling and Sun, Weixiang and Zhou, Rong and Yuan, Zhengqing and Zhang, Kai and Li, Yiwei and Liu, Tianming and Li, Quanzheng and Li, Xiang and He, Lifang and others},
  journal={arXiv preprint arXiv:2408.03286},
  year={2024}
}

Acknowledgements

We acknowledge the authors of the employed public dataset. We also thank the authors of SAM-2 and MedSAM-2 for making their valuable code publicly available. Our work is built upon MedSAM-2, which served as a critical foundation for our modifications.