|h. ddlZddlmZddlmZmZmZmZddlZ ddl Z d$de jdee dee de de jf d Zd e deeeddffd Zd e jd e de de jfdZde de j"fdZde de de dee j"fdZdee dfde de deeee ee ffdZde jdee de jfdZde jdee de jfdZd e jdee de de de jf dZde j"d e d!edee j"effd"Zd e jde jfd#Zy)%N)product)Any GeneratorListTupleboxescrop_boxorig_boxatolreturnctj|tj|j}tj|tj|j}t ||j}tj ||dddf|d}tj ||dddf|d}tj ||}tj|dS)a Determine if bounding boxes are near the edge of a cropped image region using a specified tolerance. Args: boxes (torch.Tensor): Bounding boxes in XYXY format. crop_box (List[int]): Crop box coordinates in [x0, y0, x1, y1] format. orig_box (List[int]): Original image box coordinates in [x0, y0, x1, y1] format. atol (float, optional): Absolute tolerance for edge proximity detection. Returns: (torch.Tensor): Boolean tensor indicating which boxes are near crop edges. Examples: >>> boxes = torch.tensor([[10, 10, 50, 50], [100, 100, 150, 150]]) >>> crop_box = [0, 0, 200, 200] >>> orig_box = [0, 0, 300, 300] >>> near_edge = is_box_near_crop_edge(boxes, crop_box, orig_box, atol=20.0) )dtypedeviceNr)r rtoldim)torch as_tensorfloatruncrop_boxes_xyxyisclose logical_andany)rr r r crop_box_torchorig_box_torchnear_crop_edgenear_image_edges Y/var/www/html/test/engine/venv/lib/python3.12/site-packages/ultralytics/models/sam/amg.pyis_box_near_crop_edger s*__XU[[VN__XU[[VN eX . 4 4 6E]]5.q*ASTUNmmE>$'+BTUVO&&~7GHN 99^ ++ batch_sizec' KrtfdDsJdtd|zttd|zdk7z}t|D] }Dcgc]}|||z|dz|zc}"ycc}ww)a Yield batches of data from input arguments with specified batch size for efficient processing. This function takes a batch size and any number of iterables, then yields batches of elements from those iterables. All input iterables must have the same length. Args: batch_size (int): Size of each batch to yield. *args (Any): Variable length input iterables to batch. All iterables must have the same length. Yields: (List[Any]): A list of batched elements from each input iterable. Examples: >>> data = [1, 2, 3, 4, 5] >>> labels = ["a", "b", "c", "d", "e"] >>> for batch in batch_iterator(2, data, labels): ... print(batch) [[1, 2], ['a', 'b']] [[3, 4], ['c', 'd']] [[5], ['e']] c3RK|]}t|tdk( yw)rN)len).0aargss r z!batch_iterator..@s"=1A#d1g,.=s$'z-Batched iteration must have same-size inputs.rrN)allr%intrange)r"r( n_batchesbargs ` rbatch_iteratorr0)s. C===n?nn =DG *ST!W 1Ja1O-PPI 9 KEIJcs1z>QUj$89JJKJsAB"A?7 Bmasksmask_thresholdthreshold_offsetc$|||zkDjdtjjdtj}|||z kDjdtjjdtj}||z S)a Compute the stability score for a batch of masks. The stability score is the IoU between binary masks obtained by thresholding the predicted mask logits at high and low values. Args: masks (torch.Tensor): Batch of predicted mask logits. mask_threshold (float): Threshold value for creating binary masks. threshold_offset (float): Offset applied to the threshold for creating high and low binary masks. Returns: (torch.Tensor): Stability scores for each mask in the batch. Notes: - One mask is always contained inside the other. - Memory is saved by preventing unnecessary cast to torch.int64. Examples: >>> masks = torch.rand(10, 256, 256) # Batch of 10 masks >>> mask_threshold = 0.5 >>> threshold_offset = 0.1 >>> stability_scores = calculate_stability_score(masks, mask_threshold, threshold_offset) )r)sumrint16int32)r1r2r3 intersectionsunionss rcalculate_stability_scorer;Fs2n/??@EEbPUP[P[E\``ackpkvkv`wM~(889 > >r > U Y YZ\didodo Y pF 6 !!r! n_per_sidecdd|zz }tj|d|z |}tj|dddf|df}tj|dddfd|f}tj||gdj ddS)zaGenerate a 2D grid of evenly spaced points in the range [0,1]x[0,1] for image segmentation tasks.rNr5)axis)nplinspacetilestackreshape)r<offsetpoints_one_sidepoints_xpoints_ys rbuild_point_gridrIds !j. !Fkk&!f*jAOwwtQw/*aAHwwq$w/!ZAH 88Xx(r 2 : :2q AAr!n_layersscale_per_layerc tt|dzDcgc]}tt|||zz c}Scc}w)zPGenerate point grids for multiple crop layers with varying scales and densities.r)r,rIr+)r<rJrKis rbuild_all_layer_point_gridsrNms7NST\_`T`Na b S/A!BC D bb bs!5im_size. overlap_ratioc hgg}}|\}}t||}|jdd||g|jdd}t|D]} d| dzz} t||zd| z z} ||| | } ||| | } t| Dcgc]}t| | z |z}}t| Dcgc]}t| | z |z}}t ||D]J\}}||t|| z|t|| z|g}|j||j| dzL||fScc}wcc}w)a Generate crop boxes of varying sizes for multiscale image processing, with layered overlapping regions. Args: im_size (Tuple[int, ...]): Height and width of the input image. n_layers (int): Number of layers to generate crop boxes for. overlap_ratio (float): Ratio of overlap between adjacent crop boxes. Returns: crop_boxes (List[List[int]]): List of crop boxes in [x0, y0, x1, y1] format. layer_idxs (List[int]): List of layer indices corresponding to each crop box. Examples: >>> im_size = (800, 1200) # Height, width >>> n_layers = 3 >>> overlap_ratio = 0.25 >>> crop_boxes, layer_idxs = generate_crop_boxes(im_size, n_layers, overlap_ratio) rcVttj||dz z|z|z S)zZCalculate the length of each crop given the original length, number of crops, and overlap.r)r+mathceil)orig_lenn_cropsoverlaps rcrop_lenz%generate_crop_boxes..crop_lens)499g15@GKLMMr!r>r)minappendr,r+r)rOrJrP crop_boxes layer_idxsim_him_w short_siderXi_layern_crops_per_siderWcrop_wcrop_hrM crop_box_x0 crop_box_y0x0y0boxs rgenerate_crop_boxesrirsp*  JJD$T4Jq!T4()aN?+1-mj0A8H4HIJ$ 0':$ 0':r!pointsc|\}}}}tj||gg|j}t|jdk(r|j d}||zS)zAUncrop points by adding the crop box offset to their coordinates.rkrlrrm)rrr rfrgrqrEs r uncrop_pointsrtsTLBAq \\B8*V]] ;F 6<<A!!!$ F?r!orig_horig_wc|\}}}}|dk(r|dk(r ||k(r||k(r|S|||z z |||z z } }|||z || |z f} tjjj|| dS)z]Uncrop masks by padding them to the original image size, handling coordinate transformations.r)value)rnn functionalpad) r1r rurvrfrgx1y1pad_xpad_yr{s r uncrop_masksrsNBB Qw27rV|f  R"W%vb'95E urz2urz *C 88   " "5#Q " 77r!mask area_threshmodecddl}|dvs Jd|d|dk(}||z jtj}|j |d\}}}} |dddfd d} t | D cgc]\} } | |ks | d z} } } | s|d fSdg| z}|sBt |D cgc] } | |vs|  c} xs"ttj| d zg}tj||}|d fScc} } wcc} w) a Remove small disconnected regions or holes in a mask based on area threshold and mode. Args: mask (np.ndarray): Binary mask to process. area_thresh (float): Area threshold below which regions will be removed. mode (str): Processing mode, either 'holes' to fill small holes or 'islands' to remove small disconnected regions. Returns: processed_mask (np.ndarray): Processed binary mask with small regions removed. modified (bool): Whether any regions were modified. Examples: >>> mask = np.zeros((100, 100), dtype=np.bool_) >>> mask[40:60, 40:60] = True # Create a square >>> mask[45:55, 45:55] = False # Create a hole >>> processed_mask, modified = remove_small_regions(mask, 50, "holes") rN>holesislandszProvided mode z is invalidrr5rFT) cv2astyper@uint8connectedComponentsWithStats enumerater,r+argmaxisin)rrrr correct_holes working_maskn_labelsregionsstatsrqsizesrMs small_regions fill_labelss rremove_small_regionsrs( ' 'K>${)KK 'GOM!D(00:L"%"B"Br5rrk) rnumelzerosrorr%flattenrpmaxarangerYrCrD)r1rohw in_heightrqin_height_coords bottom_edges top_edgesin_widthin_width_coords right_edges left_edges empty_filterouts rbatched_mask_to_boxrs {{5Q{{EEKK,EaE EE KKE :DAq$'JNEMM!R 8JE99U+LIq 5<<):J:J#KDRSG#TTii 0b9OL!'! z*::99-26LIq))Er*KHaa!Hq!QQOYYB7NK%hY7OIIo26MJ *, 1IJL ++z9k<Hb QC ,))"- -C+.e*q.;3;; &cr &A &Dc!fDr!)g4@)rS itertoolsrtypingrrrrnumpyr@rTensorr+rr r0r;ndarrayrIrNrirrtrstrboolrrr!rrs .. RV, <<,#'9,8