^jr<ddlmZddlmZmZmZddlZddlZddl mZ ddl m Z  d dZ d dZd dZ d ddZddd Z d dd Zy)) annotations)AnyLiteralcastN) CompactMaskctj|jd|d|dfd}|ddkrM|d }t|jd|d|dz }d}t|d|jd|dz}n4d}t|jd|d|dz }|d}|d|z|z }|ddkrM|d }t|jd|d|dz } d} t|d|jd|dz} n4d}t|jd|d|dz } |d} |d| z|z } ||kDr| |kDr|dd|| ||f|dd| | ||f<|S)a Offset the masks in an array by the specified (x, y) amount. Args: masks: 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: An array of shape `(2,)` containing int values `[dx, dy]`. Supports both positive and negative values for bidirectional movement. resolution_wh: The width and height of the desired mask resolution. Returns: Repositioned masks, optionally padded to the specified shape. Examples: ```pycon >>> 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]]]) >>> 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]]]) ``` rFN)npfullshapemin) masksoffset resolution_wh mask_arraysource_x_start source_x_enddestination_x_startdestination_x_endsource_y_start source_y_enddestination_y_startdestination_y_ends l/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/supervision/detection/utils/masks.py move_masksr sX%++a.-*:M!=+;fQi+GH  a 0%++a.6!92LM5;;q>=+;fQi+GH $Qi"1I 4~E ay1} )5;;q>=+;fQi+GH  a 0%++a.6!92LM5;;q>=+;fQi+GH $Qi"1I 4~En$)F !^L0.2MM N   1 1  1 1 3 c"ttrt}|dk(rDttj t jt jdtSt j|dft j}t|D]}j|}|j\}}tj|df}tj|df}t|j!} | dk(rddg||<}t j"||f\} } t%t j | dz|| z |z} t%t j | dz|| z |z} | | g||<ttj t j|j'tSj\}}}j!d }t j"||fdz\}}d||dk(< d fd }ddgddgf}||||z }||||z }ttj t jt j(||fj'tS) a Calculate the centroids of binary masks in a tensor. Args: masks: A 3D NumPy array of shape (num_masks, height, width). Each 2D array in the tensor represents a binary mask. Also accepts a :class:`~supervision.detection.compact_mask.CompactMask`. Returns: A 2D NumPy array of shape (num_masks, 2), where each row contains the x and y coordinates (in that order) of the centroid of the corresponding mask. r)rr dtyper r gg?)r r axiscttjtjtj ||S)N)axes)rnptNDArrayr floating tensordot)indicesr"rs r sum_over_maskz0calculate_masks_centroids..sum_over_masks-CKK ,bll5'PT.UVVr)r)npt.NDArray[np.floating]r"ztuple[list[int], list[int]]returnr+) isinstancerlenrr%r&r int_emptyintzerosfloat64rangecropr offsetssumr)floatastype column_stack)rn centroidsir5crop_hcrop_wx1y1total crop_rows crop_colscxcy _num_masksheightwidth total_pixelsvertical_indiceshorizontal_indicesr*aggregation_axis centroid_x centroid_ys` rcalculate_masks_centroidsrPZsX%% J 6 BGG,bhhvS.IJ J-/XXq!fBJJ-O q $A::a=D!ZZNFFU]]1a4()BU]]1a4()B OEz #Sz ! #%::vv.>#? Iyrvvy3567%?"DBrvvy3567%?"DB8IaL $CKK()*:*:3*?@@ % J99&9)L,.::vuo+F+L((&'L"#W)W1LW !W 78VaV4D13CD|SJ/1AB\QJ  BGGbooz:.FGNNsS rc|jtj}tj|tj tj \}}|d}|dD] }||dk7s yy)a> Checks if the binary mask contains holes (background pixels fully enclosed by foreground pixels). Args: mask: 2D binary mask where `True` indicates foreground object and `False` indicates background. Returns: True if holes are detected, False otherwise. Examples: ```pycon >>> 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" } rTF)r9r uint8cv2 findContours RETR_CCOMPCHAIN_APPROX_SIMPLE)mask mask_uint8_ hierarchyparent_contour_indexhs rcontains_holesr_snNRXX&J##J@W@WXLAy 1 A%&",  rc|dk7r|dk7r td|jtj}tj|tj }t j|||\}}t|dkDS)a Checks if the binary mask contains multiple unconnected foreground segments. Args: mask: 2D binary mask where `True` indicates foreground object and `False` indicates background. connectivity: 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: ```pycon >>> 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" } zCIncorrect connectivity value. Possible connectivity values: 4 or 8.r connectivityr ) ValueErrorr9r rT zeros_likeint32rUconnectedComponentsbool)rYrdrZlabelsnumber_of_labelsr[s rcontains_multiple_segmentsrls|dq\Q. Q  RXX&J ]]:RXX 6F11Fa  1$ %%rc|jd}|jd}t||z ||z }t||z}t||z}tjd|dz |j t}tjd|dz |j t}tj ||\} } |dd| | f} | j|jd||S)a. Resize all masks in the array to have a maximum dimension of max_dimension, maintaining aspect ratio. Args: masks: 3D array of binary masks with shape (N, H, W). max_dimension: The maximum dimension for the resized masks. Returns: Array of resized masks. r r rN)r rr1r linspacer9meshgridreshape) r max_dimension max_height max_widthscale new_height new_widthxyxvyv resized_maskss r resize_masksr|skk!nJ[[^I  *MI,E FEUZ'(JEI%&I Ay1}i077A [[A FB!R)$M  QY GGrc|jtk7r td|j\}}t j |s|j S|jtj}tj||\}} } } |dkr|j S| ddtjf} dtt j| z} |.tt j||}t||z}n| t|}n t!dt j"|t}d|| <|dk(r`| dd| | z }t j$t j&|d zd }dt j(||kd z}d||<|| S|d k(r| | k(jtj}d|z }tj*|tj,d }t/d|D]J}|| k(r | |k(}t j |s$t||j1}||ksFd||<L|| St!d)a 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: ```pycon >>> 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=3, ... mode="edge", ... connectivity=8 ... ).astype(int) 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]]) ``` The nearby 2x2 block at columns 6-7 is kept because its edge distance is within 3 pixels. The distant block at columns 9-10 is removed. zmask must be booleanrcr Nz:Either absolute_distance or relative_distance must be set.rTcentroidr r!redgerRz!mode must be 'edge' or 'centroid')r ri TypeErrorr r anycopyr9rTrUconnectedComponentsWithStats CC_STAT_AREAr1argmaxr8hypotrer2sqrtr7wheredistanceTransformDIST_L2r4r)rYabsolute_distancerelative_distancerdmoderHrIimage num_labelsrjstatsr<areas main_labeldiagonal threshold keep_labels differences distancesnearby main_maskinversedistance_transformlabel component min_distances rfilter_segments_by_distancer#sJb zzT.//JJMFE 66$<yy{ KK !E ,/+K+K L,(JyQyy{ !"c&&& 'ES5)**J$&%01+,x7  &+, UVV)+*D)IK"K  zmi &;; GGBFF;>:; RXXi945a88" F" v ! z)11"((; i- 227CKKK1j) *E "%I66)$ !3I!>!B!B!DELy(%) E" * v <==r)rnpt.NDArray[np.bool_]rznpt.NDArray[np.int32]rztuple[int, int]r,r)rznpt.NDArray[Any] | CompactMaskr,znpt.NDArray[np.int_])rYrr,ri)ra)rYrrdr1r,ri)i)rnpt.NDArray[Any]rqr1r,r)gY@Nrbr) rYrr float | Nonerrrdr1rzLiteral['edge', 'centroid']r,r) __future__rtypingrrrrUnumpyr numpy.typingr%"supervision.detection.compact_maskrrrPr_rlr|rrrrs"%% :K K !K#K K\: )::z/f67;& ;&/2;& ;&|H<',&*(. G G#G$G G & G  Gr