COCO Types

SUMMARY

Shared segmentation and return types used by the COCO datatypes.

Segmentation Types

COCORLESegmentation

TypedDict. Canonical representation of a compressed COCO Run-Length Encoding (RLE) segmentation.

This representation is used for internal storage and output.

Field	Type	Description
size	list[int]	Mask dimensions as [height, width].
counts	str	Compressed COCO RLE counts encoded as a UTF-8 string.

COCOUncompressedRLESegmentation

TypedDict. Input representation of an uncompressed COCO RLE segmentation.

The counts field contains raw run lengths instead of a compressed string. During construction, this representation is converted to COCORLESegmentation.

Field	Type	Description
size	list[int]	Mask dimensions as [height, width].
counts	list[int]	Uncompressed COCO RLE counts as non-negative integers.

COCORLESegmentationLike

Type alias:

COCORLESegmentation | COCOUncompressedRLESegmentation

Accepted input form for segmentation and segmentations constructor parameters. Both compressed and uncompressed inputs are normalized to COCORLESegmentation during construction.

COCOPolygonSegmentation

Type alias:

list[list[float]]

Canonical polygon segmentation representation. Each inner list contains the flattened coordinates of one polygon:

[x1, y1, x2, y2, ...]

A segmentation may contain one or more polygons, and each polygon must contain at least three points. This representation is used for internal storage and returned by as_polygon() and as_polygons().

COCOPolygonSegmentationLike

Type alias:

Sequence[Sequence[float] | np.ndarray]

Accepted input form for segmentation and segmentations constructor parameters. Polygon coordinates may be provided as lists, tuples, or NumPy arrays and are normalized to COCOPolygonSegmentation during construction.

Result Representations

COCOObjectDetectionResultAsMask

TypedDict. Return type of COCOObjectDetectionResult.as_mask().

Field	Type	Description
image_id	int	Source image ID.
category_id	int	Category ID.
image_height	int	Source image height.
image_width	int	Source image width.
score	float	Detection confidence score.
bbox	np.ndarray | None	(4,) float32 box in [x, y, width, height] format, or None.
segmentation	np.ndarray | None	Segmentation rasterized to an (image_height, image_width) uint8 binary mask, or None if absent.

COCOObjectDetectionResultAsPolygon

TypedDict. Return type of COCOObjectDetectionResult.as_polygon().

Field	Type	Description
image_id	int	Source image ID.
category_id	int	Category ID.
image_height	int	Source image height.
image_width	int	Source image width.
score	float	Detection confidence score.
bbox	np.ndarray | None	(4,) float32 box in [x, y, width, height] format, or None.
segmentation	COCOPolygonSegmentation | None	Segmentation extracted as polygon contours, or None if absent.

COCOObjectDetectionResultsAsMasks

TypedDict. Return type of COCOObjectDetectionResults.as_masks().

Here, N is the number of detection results.

Field	Type	Description
image_ids	np.ndarray	(N,) source image IDs.
category_ids	np.ndarray	(N,) category IDs.
image_heights	np.ndarray	(N,) source image heights.
image_widths	np.ndarray	(N,) source image widths.
scores	np.ndarray	(N,) detection confidence scores.
bboxes	np.ndarray | None	(N, 4) float32 boxes in [x, y, width, height] format, or None.
segmentations	list[np.ndarray | None]	Length-N list of uint8 binary masks, or None per entry without a segmentation.

COCOObjectDetectionResultsAsPolygons

TypedDict. Return type of COCOObjectDetectionResults.as_polygons().

Here, N is the number of detection results.

Field	Type	Description
image_ids	np.ndarray	(N,) source image IDs.
category_ids	np.ndarray	(N,) category IDs.
image_heights	np.ndarray	(N,) source image heights.
image_widths	np.ndarray	(N,) source image widths.
scores	np.ndarray	(N,) detection confidence scores.
bboxes	np.ndarray | None	(N, 4) float32 boxes in [x, y, width, height] format, or None.
segmentations	list[COCOPolygonSegmentation | None]	Length-N list of polygon segmentations, or None per entry without a segmentation.

Annotation Representations

COCOObjectDetectionAnnotationAsMask

TypedDict. Return type of COCOObjectDetectionAnnotation.as_mask().

Field	Type	Description
id	int	Annotation ID.
image_id	int	Source image ID.
category_id	int	Category ID.
bbox	np.ndarray	(4,) float32 box in [x, y, width, height] format.
area	float	Annotation area, >= 0.
iscrowd	bool	Whether the annotation represents a crowd region.
segmentation	np.ndarray | None	Segmentation rasterized to a uint8 binary mask, or None if absent.

COCOObjectDetectionAnnotationsAsMasks

TypedDict. Return type of COCOObjectDetectionAnnotations.as_masks().

Here, N is the number of annotations.

Field	Type	Description
ids	np.ndarray	(N,) annotation IDs.
image_ids	np.ndarray	(N,) source image IDs.
category_ids	np.ndarray	(N,) category IDs.
bboxes	np.ndarray	(N, 4) float32 boxes in [x, y, width, height] format.
areas	np.ndarray	(N,) annotation areas, each >= 0.
iscrowds	np.ndarray	(N,) flags indicating whether each annotation represents a crowd region.
segmentations	list[np.ndarray | None]	Length-N list of uint8 binary masks, or None per entry without a segmentation.