This project is a benchmark for general cell segmentation models. We have deployed the following 11 cell segmentation methods: MEDIAR, Cellpose, Cellpose3, SAM, Stardist, Deepcell, Cellprofiler,cellpose4,cellSAM,Hovernet(only HE),cellbin, along with code to evaluate the model segmentation performance. Through the command line or notebook, you can run 7 segmentation models or evaluate the performance of 7 models with one click.
Here is an evaluation example of DAPI staining in the DEMO:

git clone https://github.com/STOmics/cs-benchmark.git   
cd cs-benchmark 

• Create an environment for Cellpose, SAM, StarDist, and DeepCell.
• NOTE：The command does not include installing PyTorch. If you need to use GPU, please install the corresponding version of PyTorch.

# python3.8 in conda env
conda create --name=cs-benchmark python=3.8
conda activate cs-benchmark
pip install -r requirements.txt
pip install git+https://github.com/facebookresearch/segment-anything.git

• Use the following command to install the environment for mediar and cellprofiler, and add the conda path in the _py_ section of cell_seg.py or cellsegmentation_benchmark.ipynb

conda env create -f src/methods/MEDIAR/environment.yaml
conda env create -f src/methods/cellprofiler/environment.yaml
conda env create -f src/methods/hover_net/environment.yaml

• Download the necessary model file

  sam_vit_b_01ec64.pth
  MEDIAR
  Hovernet
  After downloading the model file, place it in the src/methods/models directory of the project.

zenodo download link
CNSA download link

• Modify the parameters in the following command and input it into the command line:

python cell_seg.py -i your_inputpath -o your_outputpath -m  cellpose3 sam -t ss -g True  

• Where:

• -i is the input image path
  -o is the output mask path
  -m is the algorithm(s) to be used (can specify multiple)
  -t is the image type (ss/he/dapi/mif)
  -g is the GPU index (True/False or num)

• Ensure that the images in the gt folder have filenames with "-mask" and the images in the algorithm output mask folder have filenames with "-img", with only this difference in their names.

• Modify the parameters in the following command and input it into the command line:

python src/eval/cell_eval_multi.py -g gt_path -d dt_path -o result_path

• Where:

• -g is the path to the ground truth (GT) folder
  -d is the path to the algorithm output mask folder
  -o is the output path for the results

cellsegmentation_benchmark.ipynb

if you use CellBinDB in your work, please cite it

Can Shi, Jinghong Fan, Zhonghan Deng, Huanlin Liu, Qiang Kang, Yumei Li, Jing Guo, Jingwen Wang, Jinjiang Gong, Sha Liao, Ao Chen, Ying Zhang, Mei Li, CellBinDB: a large-scale multimodal annotated dataset for cell segmentation with benchmarking of universal models, GigaScience, Volume 14, 2025, giaf069,
doi: https://doi.org/10.1101/2024.11.20.619750

cellpose
cellpose3
cellpose4
deepcell
sam
mediar
stardist
cellprofiler
Hovernet
cellSAM