CellSAM

This is the official repository for CellSAM: Segment Anything in Microscopy Images of C. elegans.

Installation

1. Create a virtual environment conda create -n cellsam python=3.10 -y and activate it conda activate cellsam
2. Install Pytorch 2.0
3. git clone https://github.com/DevoLearn/CellSAM.git
4. Enter the CellSAM folder cd CellSAM and run pip install -e .

Get Started

Download the model checkpoint and place it at e.g., work_dir/CellSAM/cellsam_vit_b

We provide three ways to quickly test the model on your microscopy images:

1. Command line

python CellSAM_Inference.py # segment the demo image

Segment other images with the following flags:

-i input_img
-o output path
--box bounding box of the segmentation target

2. Jupyter-notebook

We provide a step-by-step tutorial on CoLab

You can also run it locally with tutorial_quickstart.ipynb.

3. GUI

Install PyQt5 with pip: pip install PyQt5 or conda: conda install -c anaconda pyqt

python gui.py

Model Training

We have trained only mask decoder by freezing the image encode ,prompt encoder .

Training on multiple GPUs (Recommend)

The model was trained on one v100 gpu .

sbatch train_multi_gpus.sh

When the training process is done, please convert the checkpoint to SAM's format for convenient inference.

python utils/ckpt_convert.py # Please set the corresponding checkpoint path first

Training on one GPU

python train_one_gpu.py

Data preprocessing

Download the demo dataset and unzip it to data/CellTrain/.

This dataset contains microscopy images of C. elegans. The names of the cell label are available at CellTrackingCHALLENGE.

• Split dataset: 80% for training and 20% for testing
• Max-min normalization
• Resample image size to 1024x2014
• Save the pre-processed images and labels as png files

One can navigate through th folder for the notebooks/explore_preprocess_cell_tracking_challeneg_dataset.ipynb for converting the 3d tiff files to microscopy images to png images

Acknowledgements

• We highly appreciate all the Google summer of code organizers and dataset owners for providing the public dataset to the community.
• Thanks for the bracly alicea, mayukh deb and mainak deb through entire GSOC period
• We thank Meta AI for making the source code of segment anything publicly available.
• We also thank Alexandre Bonnet for sharing this insightful blog.
• we also thanks for the code MedSAM Code owner for there code on Medical images MedSAM

Final Report

Please find the final report for this project.