Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/abailoni/segmentation_spacem


https://github.com/abailoni/segmentation_spacem

Last synced: 24 days ago
JSON representation

Awesome Lists containing this project

README 

        
# Segmentation SpaceM

A very "early-development-stage" set of utilities to segment images using Cellpose and other tools.



## Installation

1. First, we create the conda environment. You will need PyTorch with cuda support in order to use Cellpose with GPU support. Instructions to install PyTorch could change over time, so you are welcome to double-check [here](https://pytorch.org/get-started/locally/). Since PyTorch with CUDA takes a lot of space, you may need to install the environment in a different location than the default one (for example on a scratch folder, but it will likely be deleted after a while). First, let's see the command to install it in the default location: 



   ```bash

   conda create -n segmSpaceM python=3.8 pytorch torchvision torchaudio segmfriends pytorch-cuda=12.1 -c pytorch -c nvidia -c conda-forge -c abailoni

   conda remove --force segmfriends

   conda activate segmSpaceM 

   ```



   If you want to install it in a different location, you can use the following command:



   ```bash

   conda create --prefix /path/to/folder/of/your/new/conda/env python=3.8 pytorch torchvision torchaudio segmfriends pytorch-cuda=12.1 -c pytorch -c nvidia -c conda-forge -c abailoni

   conda remove --force segmfriends

   conda activate /path/to/folder/of/your/new/conda/env

   ```



2. In order to install the package, you will need access to some SpaceM packages. Sign-in to GitLab and create a

   [personal access token](gitlab:/-/profile/personal_access_tokens) with scope `read_api`,

   or ask a SpaceM developer to provide access for you.



3. Create a file `pip.config` (or `pip.ini`) at the location specified in the

   [pip documentation](https://pip.pypa.io/en/stable/topics/configuration/#location) with the

   following content, replacing `TOKENNAME` and `TOKEN` with the respective values:



   ```ini

   [global]

   # Be aware that pip will also search in --index-url (https://pypi.org/simple/).

   # If package names clash, it will not prioritize our private repository, but the higher version number.

   extra-index-url = https://TOKENNAME:[email protected]/api/v4/groups/1245/-/packages/pypi/simple

   ```

   

4. Finally, we clone the repository and install the package:



   ```bash

   git clone https://github.com/abailoni/segmentation_spacem.git

   cd segmentation_spacem

   conda activate segmSpaceM 

   pip install -e .

   ```



## Usage

### Predicting Cellpose segmentations in batch mode

In the `example_project` directory of the repository you find an example of script and config file you may need to process some SpaceM datasets and predict cellpose segmentations for each of them. Create a copy of such folder and adapt the content to your needs:



1. In the `infer_cellpose.py` script you will need to specify the path where you want to save the results.

2. In the config file you will find comments guiding you on how to specify the paths to the images you want to process, which cellpose models you wish to run, etc.

3. You will need a machine with a GPU in order to run the `segmentation` step of the pipeline (otherwise it will take very long). The other steps can be run also without GPU support

4. Run the script with the following command:



   ```bash

   conda activate segmSpaceM 

   CUDA_VISIBLE_DEVICES=0 ipython path/to/your/script/infer_cellpose.py -- NAME_OF_THE_RUN --inherit main_v1.yml

   ```

   

    where `NAME_OF_THE_RUN` is the name you want to give to the run (it will be used to create a subdirectory in the path you specified in the `infer_cellpose.py` script), so that you can test several runs and preserve all the generated data. The `--inherit main_v1.yml` argument is used to specify the config file you want to use. You can create your own config file by copying the `main_v1.yml` file and adapting it to your needs.

5. Once you exported the results, you will find the segmentations in the `NAME_OF_THE_RUN/exported_results` subdirectory of the path you specified in the `infer_cellpose.py` script.



### Visualizing all microscopy images and segmentations in neuroglancer

1. Activate the conda environment and start jupyter notebook:



   ```bash

   conda activate segmSpaceM

   jupyter notebook

   ```



2. Use the `visualize-data-in-neuroglancer.ipynb` notebook in the `example_project` directory to start the neuroglancer server and load your results (instructions are present in the notebook).