Deploy image processing algorithms in FastAPI servers and easily run them from Napari, QuPath, and more.
The algorithm servers implement a common set of API endpoints for processing and to document the project. These include:
/process: Runs the algorithm/info: Displays documentation about the algorithm usage/sample_images: Provides one or multiple sample images
Set up and run a server with one, or multiple algorithms from the Imaging Server Kit collection (or your own):
- StarDist: Object detection with star-convex shapes
- CellPose: A generalist algorithm for cellular segmentation
- Spotiflow: Accurate and efficient spot detection
- Rembg: A tool to remove images background
- LoG detector: Laplacian of Gaussian filter
- Orientationpy: Measurement of greyscale orientations
All of these algorithm servers can be installed locally or built and run with docker. To use docker and serve multiple algorithms, the recommended way is to edit a docker-compose.yml file and pull server images from registry.rcp.epfl.ch (EPFL users only):
services:
servers_registry:
image: mallorywittwerepfl/imaging-server-kit:latest
ports:
- "8000:8000"
command: python3 start_registry.py
rembg:
image: registry.rcp.epfl.ch/imaging-server-kit/serverkit-rembg:latest
depends_on:
- servers_registry
skimage-log:
image: registry.rcp.epfl.ch/imaging-server-kit/serverkit-skimage-log:latest
depends_on:
- servers_registry
Start the server with
docker compose up
The server should be accessible at http://localhost:8000.
Build the docker images yourself
To build the algorithm server images yourself (e.g. if you are not from EPFL), see Reference Deployment.
Create a server for your project
To learn how to create a server for your project, see Algorithm Server Template.
Once the server is running, connect to it and run algorithms from Python:
from imaging_server_kit import Client
client = Client("http://localhost:8000")
print(client.algorithms)
# [`rembg`, `stardist`, `cellpose`]
algo_output = client.run_algorithm(
algorithm="rembg",
image=(...),
rembg_model_name="silueta",
)Once the server is running, use the Napari Server Kit plugin to connect to it and run algorithms in Napari.
Once the server is running, use the QuPath Extension Server Kit to connect to it and run algorithms from within QuPath.
Install the imaging-server-kit package with pip:
pip install imaging-server-kit
or clone the project and install the development version:
git clone https://github.com/Imaging-Server-Kit/imaging-server-kit.git
cd imaging-server-kit
pip install -e .
Build the imaging-server-kit docker image for a specific Python version:
docker build --build-arg PYTHON_VERSION=3.9 -t imaging-server-kit:3.9 .
Run build.sh to build docker images for Python 3.9, 3.10, 3.11, GPU and the server kit registry:
bash ./build.sh
Contributions are very welcome.
This software is distributed under the terms of the BSD-3 license.
If you encounter any problems, please file an issue along with a detailed description.