Pipeline

The Pipeline class is the orchestrator. It sequences components, manages state, handles file I/O, and coordinates background tasks.

Lifecycle

Pipeline.__init__()
    ├── Assign component IDs and output paths
    ├── Print flow chart (visualizes state/column flow)
    └── Validate pipeline (catch missing requirements early)

Pipeline.run()
    └── For each component:
        ├── Wait for any required background tasks (e.g. COCO generation)
        ├── Run component.__call__(data_state)
        └── Process result (merge GDF, save files, queue async tasks)

Flow chart

When a pipeline is constructed, it automatically prints a colored flow chart showing which state keys and GDF columns are available, produced, required, or missing at each stage. This makes it straightforward to spot configuration errors before any inference runs.

State management

All shared state lives in a single DataState object. Components read from it and return a ComponentResult. The pipeline applies the result to the state — components never mutate DataState directly.

Key state keys:

Key	Description
imagery_path	Path to the input orthomosaic
tiles_path	Path to the generated tiles directory
infer_gdf	The working GeoDataFrame of detections
infer_coco_path	Path to the current COCO annotation file
product_name	Derived from the input filename, used for output naming

Background tasks

COCO file generation is expensive and is queued as a background process after each component that produces one. The pipeline automatically waits for the relevant COCO file to finish before running any component that requires it — no manual coordination needed.

Standalone component usage

Each component can also be run individually outside a pipeline. See the Standalone Usage page for details and examples.