Cellpose Training Server

Tool to train custom cellpose models via the browser on a remote server with GPU support, given cellpose training images generated using the Cellpose Annotation Tool tool.

Installation and usage on the server with GPU support

How to build the docker image

• Clone this repository with git clone git@git.embl.de:grp-alexandrov/cellpose-training-server.git

• Move to the repository folder with cd cellpose-training-server

• To build the docker image, you will need to provide credentials to access the GitLab repository. Follow instructions here if you do not already have a pip.config file configured with your credentials. After that, you can use the tokens from your pip.conf file for the following command:

  docker build -t cellposetrainingserverapp:latest --build-arg CI_DEPLOY_USER=__your_token__ --build-arg CI_DEPLOY_PASSWORD="your_password" ./dockerfiles
  

How to start the docker image

• If you want to temporarily start the docker image, you can run the following command:

  docker run -p 8501:8501 --gpus all cellposetrainingserverapp:latest

• If you want to start the docker image and keep it running in the background (even when your server is restarted), you can run the following command:

  docker run -p 8501:8501 --gpus all --restart unless-stopped -d  cellposetrainingserverapp:latest

Afterwards, you can access the tool via the browser at http://localhost:8501