Mapper

app.infrastructure.detection.model.mtcnn_torch.mapper

View source code

Mapper for converting MTCNN (facenet-pytorch) output to DetectionResult.

See Also

• DetectionResult — normalized detection record produced by the mapper
• MTCNNTorchDetectionModel — runs the mapped model output

Layer: Infrastructure. Context: Implements concrete processing and I/O behind application-facing contracts.

MTCNNTorchDetectionMapper

Maps raw MTCNN (facenet-pytorch) detection output to DetectionResult format.

facenet_pytorch.MTCNN.detect() returns a pair (boxes, probs) where:

• boxes – numpy.ndarray of shape (N, 4) in xyxy order, or None when no face is found.
• probs – numpy.ndarray of shape (N,) with confidence scores in [0, 1], or None when no face is found.

Layer: Infrastructure. Context: Implements concrete processing or I/O behind application-facing contracts.

map(boxes: 'np.ndarray | None', probs: 'np.ndarray | None', chosen_labels: list[str] | None = None) -> list[DetectionResult] staticmethod

Map MTCNN raw inference output to a list of DetectionResult.

PARAMETER	DESCRIPTION
boxes	Array of bounding boxes in xyxy format, shape (N, 4), or None if no faces were detected. TYPE: 'np.ndarray | None'
probs	Array of confidence scores, shape (N,), or None. TYPE: 'np.ndarray | None'
chosen_labels	Optional filter; results whose label is not in this list are dropped. MTCNN always emits "person" labels. TYPE: list[str] | None DEFAULT: None

RETURNS	DESCRIPTION
list[DetectionResult]	List of DetectionResult instances filtered by confidence
list[DetectionResult]	threshold and optional label filter.