DL Detection Sorter (`f8.dl.detsorter`)¶

Sort detection payloads by a score-map SHM metric.

When to Use¶

Use f8.dl.detsorter to turn raw frame-by-frame detections into more stable ranked or identity-like output.
It is useful when the detector sees several candidates but the graph only wants one stable target or a more coherent ordering.
It is often the next step when detector output still feels too noisy for control logic.

A common chain is f8.dl.detector -> f8.dl.detsorter -> downstream logic.
If a later stage depends on continuity, sorting before tracking can improve behavior.
Compare raw detections and sorted output side by side while tuning.

A sorter cannot fully recover from poor detector quality.
Occlusion and frequent entry/exit events will still cause identity churn in hard scenes.
Decide early whether you need "best current target" or "best continuity" because that affects downstream design.

pixi run -e onnx f8pydl_detsorter

Name	Access	Required	On Node	Schema	Description
`clsWeights`	`rw`	`true`	`false`	`string / default={}`	JSON map of detection cls -> weight multiplier applied to score-map metric. Keys without a prefix are exact cls matches. Keys with 're:' prefix are Python regex patterns matched via fullmatch(). All matching rules are multiplied together. Unspecified classes default to weight 1.0. Example:
`scoreShmName`	`rw`	`true`	`true`	`string / default=`	Score-map SHM name (supports scalar1_f32 and flow2_f16).
`sortDirection`	`rw`	`true`	`true`	`string / enum[desc, asc] / default=desc`	Prefer larger scores first (desc) or smaller scores first (asc).
`scoreAggregation`	`rw`	`true`	`true`	`string / enum[mean, max, sum, median] / default=mean`	ROI reduction mode used to rank each bbox.
`lastError`	`ro`	`true`	`false`	`string / default=`	Last runtime error string (best-effort).
`active`	`rw`	`true`	`false`	`boolean / default=True`	Service lifecycle state (activate/deactivate).
`svcId`	`ro`	`true`	`false`	`string`	Readonly: current service instance id (svcId).

clsWeights (Class Weights, rw): JSON map of detection cls -> weight multiplier applied to score-map metric. Keys without a prefix are exact cls matches. Keys with 're:' prefix are Python regex patterns matched via fullmatch(). All matching rules are multiplied together. Unspecified classes default to weight 1.0. Example: {"person": 2.0, "car": 0.7, "re:^dog_.*$": 1.3} Schema: string / default={}.
scoreShmName (Score SHM, rw): Score-map SHM name (supports scalar1_f32 and flow2_f16). Schema: string / default=.
sortDirection (Sort Direction, rw): Prefer larger scores first (desc) or smaller scores first (asc). Schema: string / enum[desc, asc] / default=desc.
scoreAggregation (Score Aggregation, rw): ROI reduction mode used to rank each bbox. Schema: string / enum[mean, max, sum, median] / default=mean.
lastError (Last Error, ro): Last runtime error string (best-effort). Schema: string / default=.
active (Active, rw): Service lifecycle state (activate/deactivate). Schema: boolean / default=True.
svcId (Service Id, ro): Readonly: current service instance id (svcId). Schema: string.

None

Name	Required	On Node	Schema	Description
`detections`	`true`	`true`	`object{detections, frameId, height, model, ...}`	Detection input in schema f8visionDetections/1.

Name	Required	On Node	Schema	Description
`detections`	`true`	`true`	`object{detections, frameId, height, model, ...}`	Sorted detections in schema f8visionDetections/1.
`monitor`	`true`	`false`	`object{active, alive, cpu, error, ...}`	Unified runtime monitor snapshots (health/resource/perf/error).

None