Pop Object

The shape of a Pop, and how its top-level components run

A Pop is the pipeline definition you hand to a worker session. It is a list of components plus a few settings that apply to the whole pipeline.

Field	Type	Description
components	list	The components to run. Required. See Components.
postTransform	string	A JSONata expression applied to the result before it is returned.
defaults	object	Source settings applied to every source this Pop processes, each overridable per source. See Pop Defaults.
depthMap	object	The depth ability whose frame-level map feeds world coordinates, and whether to return the map itself. See Depth Maps.

Everything documented across these pages is what the platform accepts on the wire. Where the Python or Node SDK cannot express something, or spells it differently, the page says so.

The smallest Pop

One component, one ability:

{
    "components": [
        { "type": "inference", "ability": "eyepop.person:latest" }
    ]
}

How the component list runs

Top-level components run in parallel on the full frame. Each one sees the original media — not the output of the component before it — so list order carries no meaning.

Sequencing is expressed by forwarding, not by position in the list. A Pop that must detect a vehicle before reading its plate nests the OCR component inside the detector's forward, rather than listing the two side by side.

{
    "components": [
        { "type": "inference", "ability": "eyepop.person:latest" },
        { "type": "inference", "ability": "eyepop.text:latest" }
    ]
}

Both abilities above run against the same frame, independently, and both sets of predictions come back.

Transforming the result

postTransform reshapes the response with a JSONata expression, so your application receives the structure it wants instead of remapping the prediction itself.

Next steps

• Components — every component type and its attributes

• Forwarding — chaining components, and what each one receives

• Examples — worked pipelines end to end