Mask API

faster_coco_eval.core.mask.area(rleObjs)

Compute area of encoded masks.

Parameters:

rleObjs (Union[dict, List[dict]]) – Run-length encoding of binary mask(s).

Returns:

Area(s) of run-length encodings.

Return type:

np.ndarray

faster_coco_eval.core.mask.calculateRleForAllAnnotations(anns, img_sizes, compute_rle, compute_boundary, boundary_dilation_ratio, boundary_cpu_count)

Calculate run-length encoding for all annotations.

Parameters:

  • anns (List[dict]) – List of annotation dictionaries.

  • img_sizes (Dict[int, tuple]) – Dictionary mapping image ids to their sizes (height, width).

  • compute_rle (bool) – Whether to compute run-length encoding.

  • compute_boundary (bool) – Whether to compute boundary run-length encoding.

  • boundary_dilation_ratio (float) – Ratio of dilation to apply to the mask boundary.

  • boundary_cpu_count (int) – Number of CPUs to use for boundary computation.

Returns:

None

faster_coco_eval.core.mask.decode(rleObjs)

Decode binary masks encoded via RLE.

Parameters:

rleObjs (Union[dict, List[dict]]) – Run-length encoding of binary mask(s).

Returns:

Decoded binary mask(s).

Return type:

np.ndarray

faster_coco_eval.core.mask.encode(bimask)

Encode binary mask(s) using RLE.

Parameters:

bimask (np.ndarray) – Binary mask. Can be 2D (H, W) or 3D (H, W, N).

Returns:

Run-length encoding of the binary mask(s).

Return type:

dict

faster_coco_eval.core.mask.frPyObjects(objs, h, w)

Convert a list of objects to RLE format suitable for use in mask API.

Parameters:

  • objs (Union[ValidRleType, np.ndarray, List[float], dict]) – Objects to be converted (polygons, bboxes, etc).

  • h (int) – Height of the image.

  • w (int) – Width of the image.

Returns:

Run-length encoding of the objects.

Return type:

Union[dict, List[dict]]

faster_coco_eval.core.mask.iou(dt, gt, iscrowd)

Compute intersection over union (IoU) between two sets of run-length encoded masks.

Parameters:

  • dt (ValidRleType) – Detected masks, can be a list of RLEs or arrays.

  • gt (ValidRleType) – Ground truth masks, can be a list of RLEs or arrays.

  • iscrowd (List[int]) – List of flags indicating whether each ground truth mask is a crowd region.

Returns:

Intersection over union between dt and gt masks.

Return type:

Union[list, np.ndarray]

faster_coco_eval.core.mask.merge(rleObjs, intersect=0)

Merge a list of run-length encoded objects.

Parameters:

  • rleObjs (List[dict]) – List of run-length encoding dictionaries of binary masks.

  • intersect (int, optional) – Flag for type of merge to perform. Defaults to 0.

Returns:

Run-length encoding of merged mask.

Return type:

dict

faster_coco_eval.core.mask.rleToBoundary(rle, dilation_ratio=0.02, backend='mask_api')

Convert run-length encoding to boundary rle.

Parameters:

  • rle (dict) – Run-length encoding of a binary mask.

  • dilation_ratio (float, optional) – Ratio of dilation to apply to the mask. Defaults to 0.02.

  • backend (str, optional) – Backend to use for conversion. ‘mask_api’ uses the faster_eval_api_cpp backend, ‘opencv’ uses OpenCV for conversion. Defaults to “mask_api”.

Returns:

Run-length encoding of the boundary mask.

Return type:

dict

Raises:

ImportError – If OpenCV is selected as backend but not available.

faster_coco_eval.core.mask.rleToBoundaryCV(rle, dilation_ratio=0.02)

Convert run-length encoding to boundary rle using OpenCV backend.

Parameters:

  • rle (dict) – Run-length encoding of a binary mask.

  • dilation_ratio (float, optional) – Ratio of dilation to apply to the mask. Defaults to 0.02.

Returns:

Run-length encoding of the boundary mask.

Return type:

dict

faster_coco_eval.core.mask.segmToRle(segm, w, h)

Convert segm array to run-length encoding.

Parameters:

  • segm (Union[List[float], List[int], dict]) – Segmentation map, can be a list of floats, ints or a dictionary.

  • w (int) – Width of the image.

  • h (int) – Height of the image.

Returns:

Run-length encoding of the segmentation map.

Return type:

dict

faster_coco_eval.core.mask.toBbox(rleObjs)

Get bounding boxes surrounding encoded masks.

Parameters:

rleObjs (Union[dict, List[dict]]) – Run-length encoding of binary mask(s).

Returns:

Bounding box(es) of run-length encodings.

Return type:

np.ndarray