Segmentation masks

Mask types

There are 3 ways a segmentation mask can be encoded in the annotations json file: Polygons, RLE or COCO_RLE. Examples of what each segmentation type looks like in the JSON file:

Polygons: “segmentation”: [[510.66, 423.01, 511.72, 420.03, …, 510.45, 423.01]]
RLE: “segmentation”: {“size”: [40, 40], “counts”: [245, 5, 35, 5, …, 5, 35, 5, 1190]}
COCO_RLE: “segmentation”: {“size”: [480, 640], “counts”: “aUh2b0X…BgRU4”}

On top of those 3 segmentation types, this package introduces a fourth one called PolygonsRS. It follows the same format as the RLE and COCO_RLE types, but uses the polygons for the counts field:

PolygonsRS: “segmentation”: {“size”: [480, 640], “counts”: [[510.66, 423.01, 511.72, 420.03, …, 510.45, 423.01]]}

The advantage of this format if that he polygons can be decoded into a mask of the same as the input image without having to look up its size. However it should not be written to a json file (as it is non-standard).

Decode masks

Decode a mask to a numpy.ndarray.

Parameters:

encoded_mask (RLE | COCO_RLE | Polygons | PolygonsRS) – The encoded mask.
width (int) – Use only when the mask is a Polygon. The width of the image corresponding to the polygons.
height (int) – Use only when the mask is a Polygon. The height of the image corresponding to the polygons.

Raises:

ValueError – If the mask conversion failed.

Returns:

The decoded mask as a NumPy array.

Return type:

npt.NDArray[np.uint8]

Encode masks

rpycocotools.mask.encode(mask: npt.NDArray[np.uint8], target: Literal['polygons', 'rle', 'coco_rle', 'polygons_rs']) → Polygons | RLE | COCO_RLE | PolygonsRS:

Encode/compress a numpy.ndarray mask to the desired format.

Parameters:: mask (npt.NDArray[np.uint8]) – The uncompressed mask.
Raises:: ValueError – If the mask conversion failed.
Returns:: The compressed mask.
Return type:: Polygons | RLE | COCO_RLE | PolygonsRS

Utils

rpycocotools.mask.area(encoded_mask: RLE | COCO_RLE | PolygonsRS | Polygons) → int:

Compute the area of the given mask.

Parameters:: encoded_mask (RLE | COCO_RLE | PolygonsRS | Polygons) – The mask whose area should be computed.
Returns:: The area
Return type:: int

rpycocotools.mask.to_bbox(encoded_mask: RLE | COCO_RLE | PolygonsRS | Polygons) → rpycocotools.anns.BBox:

Compute the bounding box of the given mask.

Parameters:: encoded_mask (RLE | COCO_RLE | PolygonsRS | Polygons) – The mask whose bounding box should be computed.
Returns:: The bounding box
Return type:: BBox