InstanceDetections

inference_models.InstanceDetections dataclass

Attributes

bboxes_metadata class-attribute instance-attribute

bboxes_metadata = None

class_id instance-attribute

class_id

confidence instance-attribute

confidence

image_metadata class-attribute instance-attribute

image_metadata = None

mask instance-attribute

mask

xyxy instance-attribute

xyxy

Functions

__init__

__init__(xyxy, class_id, confidence, mask, image_metadata=None, bboxes_metadata=None)

to_supervision

to_supervision()

Convert instance segmentation detections to Supervision Detections format.

Converts the PyTorch tensor-based instance segmentation results to Supervision's NumPy-based format. This includes both bounding boxes and segmentation masks, enabling use of Supervision's mask annotators and analysis tools.

Returns:

• Detections –

  sv.Detections: Supervision Detections object with:

  • xyxy: Bounding boxes as NumPy array (N, 4) in [x1, y1, x2, y2] format

  • class_id: Class IDs as NumPy array (N,)

  • confidence: Confidence scores as NumPy array (N,)

  • mask: Segmentation masks as NumPy array (N, H, W) with boolean values

Examples:

Convert and visualize instance segmentation:

>>> import cv2
>>> import supervision as sv
>>> from inference_models import AutoModel
>>>
>>> model = AutoModel.from_pretrained("yolov8n-seg-640")
>>> image = cv2.imread("image.jpg")
>>> predictions = model(image)
>>>
>>> # Convert to Supervision format
>>> detections = predictions[0].to_supervision()
>>>
>>> # Use Supervision mask annotator
>>> mask_annotator = sv.MaskAnnotator()
>>> annotated = mask_annotator.annotate(image.copy(), detections)

Access masks:

>>> detections = predictions[0].to_supervision()
>>> print(f"Masks shape: {detections.mask.shape}")  # (N, H, W)
>>> print(f"First mask: {detections.mask[0]}")  # Boolean array

See Also

• Supervision documentation: https://supervision.roboflow.com