Plugin System

Plugins connect external algorithms (deep learning models, analysis tools) to EXACT. A plugin runs as a separate process or service and communicates with EXACT through the REST API.

How plugins work

EXACT server  ←──REST API──→  Plugin process
     │                              │
  PluginJob                    reads images,
  (queued)                     writes annotations

1. A user (or schedule) triggers a plugin job from the imageset view or via the API.
2. EXACT creates a PluginJob record with the target imageset and parameters.
3. The plugin process polls for pending jobs, processes images, and writes results back as annotations via the REST API.
4. EXACT displays the results in the annotator.

Plugin registration

Plugins are registered under Administration → Plugins. Each plugin record stores:

Field	Meaning
Name	Display name
Script	Path to the plugin entry point
Products	Which annotation types the plugin produces
Queue	Which job queue to use (default: celery)

Writing a plugin

The easiest way to write a plugin is with EXACT-Sync:

pip install EXACT-Sync

from exact_sync.v1.api import annotations_api, images_api
from exact_sync.v1.configuration import Configuration
from exact_sync.v1 import ApiClient

config = Configuration(host="http://localhost:8000")
config.username = "exact"
config.password = "exact"

with ApiClient(config) as client:
    img_api = images_api.ImagesApi(client)
    ann_api = annotations_api.AnnotationsApi(client)

    # Fetch images from an imageset
    images = img_api.images_images_list(image_set=42)

    for image in images.results:
        # Run your model
        boxes = my_model.predict(image.id)

        # Write annotations back
        for x1, y1, x2, y2, label in boxes:
            ann_api.annotations_annotations_create(body={
                "annotation_type": label_to_type_id[label],
                "image": image.id,
                "vector": {"x1": x1, "y1": y1, "x2": x2, "y2": y2},
            })

See the Inference notebook for a complete end-to-end example with a real model.

Example notebooks

Notebook	Topic
Inference Asthma	Run a trained model and upload results
Segmentation	Polygon segmentation results
AnnotationMap	Heatmap / density overlay
ClusterCells	Cluster existing annotations
PatchClassifier	Tile-level classification

All notebooks are also available via NBViewer.