Output formats and examples

Convert existing model predictions to a standard format such as COCO using one Python call. Your application runs the model; the conversion module looks up its output adapter and handles the conversion.

How to convert predictions to COCO · Runnable pipeline example

Conversion support lists importers and their evidence.

Understand the source format

A model architecture does not uniquely determine its output format. A wrapper might return boxes, write measurement tables, or save annotated photographs. The examples below show the actual documented interfaces, including units and fields that need external context. Their small values are illustrative unless explicitly labelled as a retained source example.

For the same box in a 200 × 100 image:

Representation	Values	Meaning
Pixel corners	`[10,20,50,60]`	Top-left and bottom-right corners
COCO box	`[10,20,40,40]`	Top-left position, width, height
YOLO normalized box	`[0.15,0.4,0.2,0.4]`	Center and size, divided by image dimensions
ISIR box	`[30,60,40,40]`	Pixel center and size, bottom-left origin

Each page links to its machine-readable descriptor and explains its coordinates, required context and limitations. Examples are illustrative unless labelled as retained outputs; tensor projections are not author JSON exports.