Skip to content

Masks Utils

move_masks

supervision.detection.utils.masks.move_masks(masks, offset, resolution_wh)

Offset the masks in an array by the specified (x, y) amount.

Parameters:

Name Type Description Default

masks

NDArray[bool_]

A 3D array of binary masks corresponding to the predictions. Shape: (N, H, W), where N is the number of predictions, and H, W are the dimensions of each mask.

 required

offset

NDArray[int32]

An array of shape (2,) containing int values [dx, dy]. Supports both positive and negative values for bidirectional movement.

 required

resolution_wh

Tuple[int, int]

The width and height of the desired mask resolution.

 required

Returns:

Type Description
NDArray[bool_]

(npt.NDArray[np.bool_]) repositioned masks, optionally padded to the specified shape.

Examples:

import numpy as np
import supervision as sv

mask = np.array([[[False, False, False, False],
                 [False, True,  True,  False],
                 [False, True,  True,  False],
                 [False, False, False, False]]], dtype=bool)

offset = np.array([1, 1])
sv.move_masks(mask, offset, resolution_wh=(4, 4))
# array([[[False, False, False, False],
#         [False, False, False, False],
#         [False, False,  True,  True],
#         [False, False,  True,  True]]], dtype=bool)

offset = np.array([-2, 2])
sv.move_masks(mask, offset, resolution_wh=(4, 4))
# array([[[False, False, False, False],
#         [False, False, False, False],
#         [False, False, False, False],
#         [True,  False, False, False]]], dtype=bool)
Source code in supervision/detection/utils/masks.py 
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
def move_masks(
    masks: npt.NDArray[np.bool_],
    offset: npt.NDArray[np.int32],
    resolution_wh: tuple[int, int],
) -> npt.NDArray[np.bool_]:
    """
    Offset the masks in an array by the specified (x, y) amount.

    Args:
        masks (npt.NDArray[np.bool_]): A 3D array of binary masks corresponding to the
            predictions. Shape: `(N, H, W)`, where N is the number of predictions, and
            H, W are the dimensions of each mask.
        offset (npt.NDArray[np.int32]): An array of shape `(2,)` containing int values
            `[dx, dy]`. Supports both positive and negative values for bidirectional
            movement.
        resolution_wh (Tuple[int, int]): The width and height of the desired mask
            resolution.

    Returns:
        (npt.NDArray[np.bool_]) repositioned masks, optionally padded to the specified
            shape.

    Examples:
        ```python
        import numpy as np
        import supervision as sv

        mask = np.array([[[False, False, False, False],
                         [False, True,  True,  False],
                         [False, True,  True,  False],
                         [False, False, False, False]]], dtype=bool)

        offset = np.array([1, 1])
        sv.move_masks(mask, offset, resolution_wh=(4, 4))
        # array([[[False, False, False, False],
        #         [False, False, False, False],
        #         [False, False,  True,  True],
        #         [False, False,  True,  True]]], dtype=bool)

        offset = np.array([-2, 2])
        sv.move_masks(mask, offset, resolution_wh=(4, 4))
        # array([[[False, False, False, False],
        #         [False, False, False, False],
        #         [False, False, False, False],
        #         [True,  False, False, False]]], dtype=bool)
        ```
    """
    mask_array = np.full((masks.shape[0], resolution_wh[1], resolution_wh[0]), False)

    if offset[0] < 0:
        source_x_start = -offset[0]
        source_x_end = min(masks.shape[2], resolution_wh[0] - offset[0])
        destination_x_start = 0
        destination_x_end = min(resolution_wh[0], masks.shape[2] + offset[0])
    else:
        source_x_start = 0
        source_x_end = min(masks.shape[2], resolution_wh[0] - offset[0])
        destination_x_start = offset[0]
        destination_x_end = offset[0] + source_x_end - source_x_start

    if offset[1] < 0:
        source_y_start = -offset[1]
        source_y_end = min(masks.shape[1], resolution_wh[1] - offset[1])
        destination_y_start = 0
        destination_y_end = min(resolution_wh[1], masks.shape[1] + offset[1])
    else:
        source_y_start = 0
        source_y_end = min(masks.shape[1], resolution_wh[1] - offset[1])
        destination_y_start = offset[1]
        destination_y_end = offset[1] + source_y_end - source_y_start

    if source_x_end > source_x_start and source_y_end > source_y_start:
        mask_array[
            :,
            destination_y_start:destination_y_end,
            destination_x_start:destination_x_end,
        ] = masks[:, source_y_start:source_y_end, source_x_start:source_x_end]

    return mask_array

contains_holes

supervision.detection.utils.masks.contains_holes(mask)

Checks if the binary mask contains holes (background pixels fully enclosed by foreground pixels).

Parameters:

Name Type Description Default

mask

NDArray[bool_]

2D binary mask where True indicates foreground object and False indicates background.

 required

Returns:

Type Description
bool

True if holes are detected, False otherwise.

Examples:

import numpy as np
import supervision as sv

mask = np.array([
    [0, 0, 0, 0, 0],
    [0, 1, 1, 1, 0],
    [0, 1, 0, 1, 0],
    [0, 1, 1, 1, 0],
    [0, 0, 0, 0, 0]
]).astype(bool)

sv.contains_holes(mask=mask)
# True

mask = np.array([
    [0, 0, 0, 0, 0],
    [0, 1, 1, 1, 0],
    [0, 1, 1, 1, 0],
    [0, 1, 1, 1, 0],
    [0, 0, 0, 0, 0]
]).astype(bool)

sv.contains_holes(mask=mask)
# False

contains_holes

Source code in supervision/detection/utils/masks.py 
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
def contains_holes(mask: npt.NDArray[np.bool_]) -> bool:
    """
    Checks if the binary mask contains holes (background pixels fully enclosed by
    foreground pixels).

    Args:
        mask (npt.NDArray[np.bool_]): 2D binary mask where `True` indicates foreground
            object and `False` indicates background.

    Returns:
        True if holes are detected, False otherwise.

    Examples:
        ```python
        import numpy as np
        import supervision as sv

        mask = np.array([
            [0, 0, 0, 0, 0],
            [0, 1, 1, 1, 0],
            [0, 1, 0, 1, 0],
            [0, 1, 1, 1, 0],
            [0, 0, 0, 0, 0]
        ]).astype(bool)

        sv.contains_holes(mask=mask)
        # True

        mask = np.array([
            [0, 0, 0, 0, 0],
            [0, 1, 1, 1, 0],
            [0, 1, 1, 1, 0],
            [0, 1, 1, 1, 0],
            [0, 0, 0, 0, 0]
        ]).astype(bool)

        sv.contains_holes(mask=mask)
        # False
        ```

    ![contains_holes](https://media.roboflow.com/supervision-docs/contains-holes.png){ align=center width="800" }
    """  # noqa E501 // docs
    mask_uint8 = mask.astype(np.uint8)
    _, hierarchy = cv2.findContours(mask_uint8, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)

    if hierarchy is not None:
        parent_contour_index = 3
        for h in hierarchy[0]:
            if h[parent_contour_index] != -1:
                return True
    return False

contains_multiple_segments

supervision.detection.utils.masks.contains_multiple_segments(mask, connectivity=4)

Checks if the binary mask contains multiple unconnected foreground segments.

Parameters:

Name Type Description Default

mask

NDArray[bool_]

2D binary mask where True indicates foreground object and False indicates background.

 required

connectivity (int)

Default: 4 is 4-way connectivity, which means that foreground pixels are the part of the same segment/component if their edges touch. Alternatively: 8 for 8-way connectivity, when foreground pixels are connected by their edges or corners touch.

 required

Returns:

Type Description
bool

True when the mask contains multiple not connected components, False otherwise.

Raises:

Type Description
ValueError

If connectivity(int) parameter value is not 4 or 8.

Examples:

import numpy as np
import supervision as sv

mask = np.array([
    [0, 0, 0, 0, 0, 0],
    [0, 1, 1, 0, 1, 1],
    [0, 1, 1, 0, 1, 1],
    [0, 0, 0, 0, 0, 0],
    [0, 1, 1, 1, 0, 0],
    [0, 1, 1, 1, 0, 0]
]).astype(bool)

sv.contains_multiple_segments(mask=mask, connectivity=4)
# True

mask = np.array([
    [0, 0, 0, 0, 0, 0],
    [0, 1, 1, 1, 1, 1],
    [0, 1, 1, 1, 1, 1],
    [0, 1, 1, 1, 1, 1],
    [0, 1, 1, 1, 1, 1],
    [0, 0, 0, 0, 0, 0]
]).astype(bool)

sv.contains_multiple_segments(mask=mask, connectivity=4)
# False

contains_multiple_segments

Source code in supervision/detection/utils/masks.py 
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
def contains_multiple_segments(
    mask: npt.NDArray[np.bool_], connectivity: int = 4
) -> bool:
    """
    Checks if the binary mask contains multiple unconnected foreground segments.

    Args:
        mask (npt.NDArray[np.bool_]): 2D binary mask where `True` indicates foreground
            object and `False` indicates background.
        connectivity (int) : Default: 4 is 4-way connectivity, which means that
            foreground pixels are the part of the same segment/component
            if their edges touch.
            Alternatively: 8 for 8-way connectivity, when foreground pixels are
            connected by their edges or corners touch.

    Returns:
        True when the mask contains multiple not connected components, False otherwise.

    Raises:
        ValueError: If connectivity(int) parameter value is not 4 or 8.

    Examples:
        ```python
        import numpy as np
        import supervision as sv

        mask = np.array([
            [0, 0, 0, 0, 0, 0],
            [0, 1, 1, 0, 1, 1],
            [0, 1, 1, 0, 1, 1],
            [0, 0, 0, 0, 0, 0],
            [0, 1, 1, 1, 0, 0],
            [0, 1, 1, 1, 0, 0]
        ]).astype(bool)

        sv.contains_multiple_segments(mask=mask, connectivity=4)
        # True

        mask = np.array([
            [0, 0, 0, 0, 0, 0],
            [0, 1, 1, 1, 1, 1],
            [0, 1, 1, 1, 1, 1],
            [0, 1, 1, 1, 1, 1],
            [0, 1, 1, 1, 1, 1],
            [0, 0, 0, 0, 0, 0]
        ]).astype(bool)

        sv.contains_multiple_segments(mask=mask, connectivity=4)
        # False
        ```

    ![contains_multiple_segments](https://media.roboflow.com/supervision-docs/contains-multiple-segments.png){ align=center width="800" }
    """  # noqa E501 // docs
    if connectivity != 4 and connectivity != 8:
        raise ValueError(
            "Incorrect connectivity value. Possible connectivity values: 4 or 8."
        )
    mask_uint8 = mask.astype(np.uint8)
    labels = np.zeros_like(mask_uint8, dtype=np.int32)
    number_of_labels, _ = cv2.connectedComponents(
        mask_uint8, labels, connectivity=connectivity
    )
    return number_of_labels > 2

filter_segments_by_distance

supervision.detection.utils.masks.filter_segments_by_distance(mask, absolute_distance=100.0, relative_distance=None, connectivity=8, mode='edge')

Keep the largest connected component and any other components within a distance threshold.

Distance can be absolute in pixels or relative to the image diagonal.

Parameters:

Name Type Description Default

mask

NDArray[bool_]

Boolean mask HxW.

 required

absolute_distance

float | None

Max allowed distance in pixels to the main component. Ignored if relative_distance is provided.

 100.0

relative_distance

float | None

Fraction of the diagonal. If set, threshold = fraction * sqrt(H^2 + W^2).

 None

connectivity

int

Defines which neighboring pixels are considered connected. - 4-connectedness: Only orthogonal neighbors. 

[ ][X][ ]
[X][O][X]
[ ][X][ ]
- 8-connectedness: Includes diagonal neighbors. 
[X][X][X]
[X][O][X]
[X][X][X]
Default is 8.

 8

mode

Literal['edge', 'centroid']

Defines how distance between components is measured. - "edge": Uses distance between nearest edges (via distance transform). - "centroid": Uses distance between component centroids.

 'edge'

Returns:

Type Description
NDArray[bool_]

Boolean mask after filtering.

Examples:

import numpy as np
import supervision as sv

mask = np.array([
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
    [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
    [0, 1, 1, 1, 0, 0, 1, 1, 0, 0, 0, 0],
    [0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
], dtype=bool)

sv.filter_segments_by_distance(
    mask,
    absolute_distance=2,
    mode="edge",
    connectivity=8
).astype(int)

# np.array([
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 1, 1, 1, 0, 0, 1, 1, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
# ], dtype=bool)

# The nearby 2×2 block at columns 6–7 is kept because its edge distance
# is within 2 pixels. The distant block at columns 9-10 is removed.
Source code in supervision/detection/utils/masks.py 
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
def filter_segments_by_distance(
    mask: npt.NDArray[np.bool_],
    absolute_distance: float | None = 100.0,
    relative_distance: float | None = None,
    connectivity: int = 8,
    mode: Literal["edge", "centroid"] = "edge",
) -> npt.NDArray[np.bool_]:
    """
    Keep the largest connected component and any other components within a distance
    threshold.

    Distance can be absolute in pixels or relative to the image diagonal.

    Args:
        mask: Boolean mask HxW.
        absolute_distance: Max allowed distance in pixels to the main component.
            Ignored if `relative_distance` is provided.
        relative_distance: Fraction of the diagonal. If set, threshold = fraction * sqrt(H^2 + W^2).
        connectivity: Defines which neighboring pixels are considered connected.
            - 4-connectedness: Only orthogonal neighbors.
              ```
              [ ][X][ ]
              [X][O][X]
              [ ][X][ ]
              ```
            - 8-connectedness: Includes diagonal neighbors.
              ```
              [X][X][X]
              [X][O][X]
              [X][X][X]
              ```
            Default is 8.
        mode: Defines how distance between components is measured.
            - "edge": Uses distance between nearest edges (via distance transform).
            - "centroid": Uses distance between component centroids.

    Returns:
        Boolean mask after filtering.

    Examples:
        ```python
        import numpy as np
        import supervision as sv

        mask = np.array([
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
            [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
            [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
            [0, 1, 1, 1, 0, 0, 1, 1, 0, 0, 0, 0],
            [0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0],
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0],
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0],
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
            [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        ], dtype=bool)

        sv.filter_segments_by_distance(
            mask,
            absolute_distance=2,
            mode="edge",
            connectivity=8
        ).astype(int)

        # np.array([
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 1, 1, 1, 0, 0, 1, 1, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        #     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        # ], dtype=bool)

        # The nearby 2×2 block at columns 6–7 is kept because its edge distance
        # is within 2 pixels. The distant block at columns 9-10 is removed.
        ```
    """  # noqa E501 // docs
    if mask.dtype != bool:
        raise TypeError("mask must be boolean")

    height, width = mask.shape
    if not np.any(mask):
        return mask.copy()

    image = mask.astype(np.uint8)
    num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(
        image, connectivity=connectivity
    )

    if num_labels <= 1:
        return mask.copy()

    areas = stats[1:, cv2.CC_STAT_AREA]
    main_label = 1 + int(np.argmax(areas))

    if relative_distance is not None:
        diagonal = float(np.hypot(height, width))
        threshold = float(relative_distance) * diagonal
    else:
        threshold = float(absolute_distance)

    keep_labels = np.zeros(num_labels, dtype=bool)
    keep_labels[main_label] = True

    if mode == "centroid":
        differences = centroids[1:] - centroids[main_label]
        distances = np.sqrt(np.sum(differences**2, axis=1))
        nearby = 1 + np.where(distances <= threshold)[0]
        keep_labels[nearby] = True
    elif mode == "edge":
        main_mask = (labels == main_label).astype(np.uint8)
        inverse = 1 - main_mask
        distance_transform = cv2.distanceTransform(inverse, cv2.DIST_L2, 3)
        for label in range(1, num_labels):
            if label == main_label:
                continue
            component = labels == label
            if not np.any(component):
                continue
            min_distance = float(distance_transform[component].min())
            if min_distance <= threshold:
                keep_labels[label] = True
    else:
        raise ValueError("mode must be 'edge' or 'centroid'")

    return keep_labels[labels]

Comments