3j%SSKJr SSKJrJrJr SSKrSSKJr SSKJ r SSK J r SSK Jr SSK Jr /S Qr"S S \5r"S S 5r"SS\R(5rg))Enum)DictListOptionalN)nn)nms)YuNet)PriorBox)decode) FaceDetectorFaceDetectorResult FaceKeypointc,\rSrSrSrSrSrSrSrSr Sr g ) r zcDefine the keypoints detected in a face. The left/right convention is based on the screen viewer. rN) __name__ __module__ __qualname____firstlineno____doc__EYE_LEFT EYE_RIGHTNOSE MOUTH_LEFT MOUTH_RIGHT__static_attributes__rW/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/contrib/face_detection.pyrr s" HI DJKr!rc\rSrSrSrS\R SS4SjrSS\\RS\\RSS4S jjr \ S\R 4S j5r \ S\R 4S j5r\ S\R 4S j5r\ S\R 4S j5rS\S\R 4Sjr\ S\R 4Sj5r\ S\R 4Sj5r\ S\R 4Sj5r\ S\R 4Sj5r\ S\R 4Sj5r\ S\R 4Sj5r\ S\R 4Sj5rSrg)r -zEncapsulate the results obtained by the :py:class:`kornia.contrib.FaceDetector`. Args: data: the encoded results coming from the feature detector with shape :math:`(14,)`. datareturnNc`[U5S:a[SURS35eXlg)Nz.Result must comes as vector of size(14). Got: .)len ValueErrorshape_data)selfr%s r"__init__FaceDetectorResult.__init__5s, t9r>Mdjj\YZ[\ \ r!devicedtypecBURRXS9UlU$)z)Like :func:`torch.nn.Module.to()` method.r1r2)r-to)r.r1r2s r"r5FaceDetectorResult.to:sZZ]]&]>  r!c URS$)z'The bounding box top-left x-coordinate..rr-r.s r"xminFaceDetectorResult.xmin?zz&!!r!c URS$)z'The bounding box top-left y-coordinate..rr9r:s r"yminFaceDetectorResult.yminDr=r!c URS$)z+The bounding box bottom-right x-coordinate.).rr9r:s r"xmaxFaceDetectorResult.xmaxIr=r!c URS$)z+The bounding box bottom-right y-coordinate.).rr9r:s r"ymaxFaceDetectorResult.ymaxNr=r!keypointcU[R:XaURSnU$U[R:XaURSnU$U[R:XaURSnU$U[R :XaURSnU$U[R :XaURSnU$[SUS35e)zsGet the [x y] position of a given facial keypoint. Args: keypoint: the keypoint type to return the position. ).)r).)).) ).) ).) zNot valid keypoint type. Got: r))rrr-rrrrr+)r.rHouts r" get_keypointFaceDetectorResult.get_keypointSs |,, ,**[)C // /**[)C ** ***[)C 00 0**]+C  11 1**]+C =hZqIJ Jr!c URS$)zThe detection score.).r9r:s r"scoreFaceDetectorResult.scorehszz'""r!c4URUR- $)zThe bounding box width.)rCr;r:s r"widthFaceDetectorResult.widthmyy499$$r!c4URUR- $)zThe bounding box height.)rFr@r:s r"heightFaceDetectorResult.heightrr]r!c URS$)BThe [x y] position of the top-left coordinate of the bounding box.).)rrr9r:s r"top_leftFaceDetectorResult.top_leftwzz+&&r!cLURnUS==UR- ss'U$)rbr8)rcr[r.rSs r" top_rightFaceDetectorResult.top_right|s$mm F tzz!  r!c URS$)zFThe [x y] position of the bottom-right coordinate of the bounding box.).)rrr9r:s r" bottom_rightFaceDetectorResult.bottom_rightrer!cLURnUS==UR- ss'U$)rbr?)rcr_rgs r" bottom_leftFaceDetectorResult.bottom_lefts$mm F t{{"  r!r9)NN)rrrrrtorchTensorr/rr1r2r5propertyr;r@rCrFrrTrXr[r_rcrhrkrnr rr!r"r r -sU\\d %,,/x ?T`t "ell"""ell"""ell"""ell""\ell*#u||##%u||%%% %%'%,,''5<< 'ell''U\\r!r c ^\rSrSrSrSS\S\S\S\SS4 U4S jjjrS \RS\R4S jr S \ \ \R4S \S\S\ \R4SjrS \RS\ \R4SjrSrU=r$)r aDetect faces in a given image using YuNet model. This is a high-level API that wraps the :py:class:`kornia.models.YuNet` model for face detection. By default, it uses the method described in :cite:`facedetect-yu`. Args: top_k: the maximum number of detections to return before the nms. confidence_threshold: the threshold used to discard detections. nms_threshold: the threshold used by the nms for iou. keep_top_k: the maximum number of detections to return after the nms. Return: A list of B tensors with shape :math:`(N,15)` to be used with :py:class:`kornia.contrib.FaceDetectorResult`. Example: >>> img = torch.rand(1, 3, 320, 320) >>> detect = FaceDetector() >>> res = detect(img) top_kconfidence_threshold nms_threshold keep_top_kr&Nc>[TU]5 XlX lX0lX@lS/SQSS/SS//SQ//SQS S /S S .Ul/SQSS/SS//SQ/Ul/SQUlS S /Ul S Ul [S SS9Ul [Ulg)Nr )rOr0@`))rMrzrr}g?g?F)name min_sizesstepsvariancecliptestT) pretrained)superr/rurvrwrxconfigrrrrr model nms_korniar)r.rurvrwrx __class__s r"r/FaceDetector.__init__s  $8!*$&R2r(OL$c    'R2r(OL$ c   6d3 r!imagecU$)aPreprocess input images before feeding them to YuNet. Args: image: Batch of RGB images with shape :math:`(B, 3, H, W)`, where: - ``B`` is batch size, - ``3`` is the RGB channel dimension, - ``H`` is image height, - ``W`` is image width. Returns: The preprocessed image tensor. The current implementation returns the input unchanged and acts as an extension point for custom pipelines. r)r.rs r" preprocessFaceDetector.preprocesss  r!r%r_r[cUSUSUSpen[R"X2X2X2X2X2X2X2/URURS9n[ UR UR URX#4S9nURURUR5n/n [URS5GHn [XJU"5UR5n X-n XZSS2S4XjSS2S4pXRS S 5-R5nXR:nXXpUR!S S 9SSUR"nU UUUp[R$"XSS2S44S S9nUR'U SS2SS24XR(5n[+U5S:a UUSS24nU R-USUR.5 GM U $)awDecode model outputs into filtered face detections. Args: data: Raw output dictionary from YuNet containing ``loc``, ``conf``, and ``iou`` tensors. height: Input image height used to scale decoded box coordinates. width: Input image width used to scale decoded box coordinates. Returns: A list with one tensor per batch element. Each tensor is shaped :math:`(N, 15)`, where ``N`` is the number of detections kept after confidence filtering and non-maximum suppression (NMS). The 15 values per detection are: - 14 geometry values (bounding box + five keypoints), - 1 confidence score. locconfiour4) image_sizerNrgg?T) descending)dimr)rptensorr1r2r rrrr5ranger,_decoderclampsqrtrvsortrucatrrwr*appendrx)r.r%r_r[rrrscalepriors batched_dets batch_elemboxes cls_scores iou_scoresscoresindsorderdetskeeps r" postprocessFaceDetector.postprocesss$ed6lDK3 E5%X]gl u::))  $..$**diiVO\3::syy1+-  ! -JCOVXt}}EEME&*a*:%;SQPQAQ=R  #3#3C#==CCEF555D!K6KK4K03LdjjAE!%L&-699eAtG_52>D88E!RaR%L&2D2DED4y1}D!G}   %6t 7 81.2r!cURU5nURU5nURX2RSURS5$)zDetect faces in a given batch of images. Args: image: batch of images :math:`(B,3,H,W)` Return: List[torch.Tensor]: list with the boxes found on each image. :math:`Bx(N,15)`. r)rrrr,)r.rimgrSs r"forwardFaceDetector.forwardsBooe$jjoYYr]CIIbMBBr!) rrvrrxrrrrwrrur)i333333?ri)rrrrrintfloatr/rprqrrstrrrrr __classcell__)rs@r"r r s,sv7<SXlo ,  7S%,,%6 777S7UYZ_ZfZfUg7r CU\\ Cd5<<.@ C Cr!r )enumrtypingrrrrprkornia.geometry.bboxrrkornia.models.yunetr kornia.models.yunet.processorsr r r__all__rr Moduler rr!r"rsT$'' 2%3< @ 4 ``FAC299ACr!