Detect Objects (Ultralytics YOLO)

How it fits

Typical backends

Caveats

• I/O contractThe output is a oneof resolved at deploy time by the connected sink, and every shape is a projection of one per-detection record so the shapes agree detection-for-detection: [BoundingBox] is box geometry plus class, [Rectangle<Double>] is box geometry only, [DetectedClass] is class id plus confidence only, [Segmentation] is a per-instance mask plus class, [Mask] is the per-instance mask only, UInt64 is the detection count, and Image is the source frame with boxes drawn.
• Hard constraintThe bound weights directory must contain exactly one `.pt` checkpoint; zero or more than one aborts startup. Resolving the output to either mask shape — [Segmentation] or [Mask] — requires a YOLO-seg checkpoint, otherwise startup aborts with a task mismatch.
• Parameter interactionthreshold is the base confidence floor, class_thresholds overrides it per class id, and object_classes is an allow-list passed to the model (an empty list keeps every class). iou bounds NMS overlap.
• Accuracyimg_size should match the training-time input size; a mismatch silently degrades accuracy. The emitted class id space is the checkpoint's own, so downstream class-id logic breaks if the weights change.
• FallbackA frame that decodes empty yields the safe empty value for the resolved arm (an empty list, a zero count, or a blank overlay frame). device starting with `cuda` silently falls back to CPU when CUDA is unavailable.
• State lifetimeAll configuration keys and the loaded weights are captured once at startup; runtime changes have no effect and require a redeploy.