Segmentation Server

This project provides a REST API for DL segmentation methods that automatically manages installtion on-demand based on mlflow-configured git repositories. The REST API is deployed in a docker container featuring:

• nvidia gpu support
• conda/mamba
• base configuration in "serve" environment

Setup

You can easily launch the server locally using

docker build -t segserve .
docker run -it -p 8000:8000 segserve

This launches the segmentation server at port 8000.

Hint: For using the GPU inside the container you usually have to launch it using

docker build -t segserve .
docker run -it -p 8000:8000 --gpus all segserve

or use the runtime: nvidia option in docker-compose (use a recent docker-compose version).

Usage

To use SegServe for segmentation you can access the REST API, for example, using curl. Here is a bash example that first downloads an example image and then requests the segmentation from SegServe using the [Omnipose] segmentation method:

wget https://fz-juelich.sciebo.de/s/wAXbC0MoN1G3ST7/download -O localhost.jpeg
curl -X 'POST' \
  'http://localhost:8001/batch-image-prediction/?repo=https%3A%2F%2Fgithub.com%2Fhip-satomi%2FCellpose-Executor.git&entry_point=omnipose&version=main' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'files=@localhost.jpeg;type=image/jpeg'

Custom DLS Executor

You can include costum implemented DLS methods into segServe. Read this tutorial to build your own git repository capable of running DLS in segServe.

API documentation

The server comes with an api documentation that is accessible at localhost:8000/docs. It also provides an interactive user interface to execute various segmentation approaches with different parameters and image files.

Running behind a proxy

For correctly running behind a proxy you need to a the --root-path command to your uvicorn launch specifying the sub-url location where your seg-serve instance is listening. For example

docker run ... uvicorn uvicorn --host 0.0.0.0 --root-path=/this/is/your/sub-url/ main:app

If you do not configure this, the endpoints will still be accessible but docs interactive usage will not work.

Security

Please note that this REST API allows to execute arbitrary git repositories. Therefore, usage is only recommended with trusted users!

Debugging

For debugging we recommend to run SegServe directly with uvicorn:

uvicorn --host 0.0.0.0 main:app