FPhvddlZddlZddlmZddlZddlZddlZddlm Z ddl m Z m Z ddlmZddlmZddlmZddlmZdd lmZmZGd d ZGd d ZGddZGddeZGddeZGddZGddZGddZ GddZ!GddZ"GddZ#Gd d!Z$d,d"Z%d-d#Z&d$Z' d.d%Z(Gd&d'Z)Gd(d)Z*Gd*d+Z+y)/N)deepcopy)LOGGERcolorstr) check_version) Instances)bbox_ioa) segment2box)polygons2maskspolygons2masks_overlapc0eZdZdZddZdZdZdZdZy) BaseTransformap Base class for image transformations. This is a generic transformation class that can be extended for specific image processing needs. The class is designed to be compatible with both classification and semantic segmentation tasks. Methods: __init__: Initializes the BaseTransform object. apply_image: Applies image transformation to labels. apply_instances: Applies transformations to object instances in labels. apply_semantic: Applies semantic segmentation to an image. __call__: Applies all label transformations to an image, instances, and semantic masks. Ncy)z%Initializes the BaseTransform object.Nselfs cC:\Users\daisl\Desktop\realtime-object-detection\venv\Lib\site-packages\ultralytics/data/augment.py__init__zBaseTransform.__init__% cy)z(Applies image transformations to labels.Nrrlabelss r apply_imagezBaseTransform.apply_image)rrcy)z6Applies transformations to object instances in labels.Nrrs rapply_instanceszBaseTransform.apply_instances-rrcy)z*Applies semantic segmentation to an image.Nrrs rapply_semanticzBaseTransform.apply_semantic1rrcj|j||j||j|y)zMApplies all label transformations to an image, instances, and semantic masks.N)rrrrs r__call__zBaseTransform.__call__5s,   V$ F#rreturnN) __name__ __module__ __qualname____doc__rrrrr rrrrrs      $rrc.eZdZdZdZdZdZdZdZy)Composez3Class for composing multiple image transformations.c||_y)z9Initializes the Compose object with a list of transforms.N transforms)rr+s rrzCompose.__init__?s $rc8|jD] }||} |S)z2Applies a series of transformations to input data.r*)rdatats rr zCompose.__call__CsAT7D! rc:|jj|y)z;Appends a new transform to the existing list of transforms.N)r+append)r transforms rr0zCompose.appendIs y)rc|jS)z:Converts the list of transforms to a standard Python list.r*rs rtolistzCompose.tolistMs rc|jjddj|jDcgc]}|c}dScc}w)z.Returns a string representation of the object.(, )) __class__r#joinr+)rr.s r__repr__zCompose.__repr__QsE..))*!DIIt6W!!v6W,X+YYZ[[6Ws A N) r#r$r%r&rr r0r3r:rrrr(r(<s=% *\rr(c,eZdZdZdddZdZdZdZy) BaseMixTransformze Class for base mix (MixUp/Mosaic) transformations. This implementation is from mmyolo. Nc.||_||_||_y)zUInitializes the BaseMixTransform object with dataset, pre_transform, and probability.Ndataset pre_transformp)rr?r@rAs rrzBaseMixTransform.__init__]s *rctjdd|jkDr|S|j}t |t r|g}|Dcgc]}|j j|}}|j't|D]\}}|j|||<||d<|j|}|jdd|Scc}w)zMApplies pre-processing transforms and mixup/mosaic transforms to labels data.rr N mix_labels) randomuniformrA get_indexes isinstanceintr?get_image_and_labelr@ enumerate_mix_transformpop)rrindexesirCr-s rr zBaseMixTransform.__call__cs >>!Q $&& (M""$ gs #iGDKK7adll66q97 K    )$Z04 $ 2 24 8 1 1)|$$V, <& Ls "Cct)z=Applies MixUp or Mosaic augmentation to the label dictionary.NotImplementedErrorrs rrKzBaseMixTransform._mix_transformz!!rct)z8Gets a list of shuffled indexes for mosaic augmentation.rPrs rrFzBaseMixTransform.get_indexes~rRrNr!)r#r$r%r&rr rKrFrrrr<r<Vs  .""rr<cTeZdZdZd fd Zd dZdZdZdZe dZ dZ xZ S) Mosaicaj Mosaic augmentation. This class performs mosaic augmentation by combining multiple (4 or 9) images into a single mosaic image. The augmentation is applied to a dataset with a given probability. Attributes: dataset: The dataset on which the mosaic augmentation is applied. imgsz (int, optional): Image size (height and width) after mosaic pipeline of a single image. Default to 640. p (float, optional): Probability of applying the mosaic augmentation. Must be in the range 0-1. Default to 1.0. n (int, optional): The grid size, either 4 (for 2x2) or 9 (for 3x3). cd|cxkrdks nJd|d|dvsJdt|||||_||_| dz| dzf|_||_y ) zKInitializes the object with a dataset, image size, probability, and border.r?z3The probability should be in range [0, 1], but got .) zgrid must be equal to 4 or 9.)r?rAN)superrr?imgszbordern)rr?r_rArar8s rrzMosaic.__init__svA}}X STUSVVWXX}F{;;;{ A.  v{UFaK0 rc 2|rAtjt|jj|j dz St |j dz Dcgc].}tjdt|jdz 0c}Scc}w)z1Return a list of random indexes from the dataset.r kr) rDchoiceslistr?bufferrarangerandintlen)rrg_s rrFzMosaic.get_indexessn >>$t||':':";tvvzJ JFKDFFUVJFWXFWFNN1c$,,&7!&;z"Mosaic._mosaic4..s,Jk#fnnaRQ34ks03r[rrCr img resized_shaper]rdtype) r_r`rhrLnpfullshapeuint8maxmin_update_labelsr0 _cat_labels)rr mosaic_labelsycxcrN labels_patchrwhwimg4x1ay1ax2ay2ax1by1bx2by2bpadwpadh final_labelsrus @rrozMosaic._mosaic4s JJJdkkJBqA%&!V6 1Ea!e1LLu%C##O4DAqAvwwAq1uciil;SQ%(a^Sa^R%K"S#s%&#)_a39oq!%K"S#sa%'R!VQR!VQU9KR%O"S#s%&S3YQc 9JA%M"S#sa%(a^RSQQ=O%O"S#s%&#)_aCc 1O%O"S#s%(S#c')9%:DS#c'! "9D9D..|T4HL   .56'' 6 " Urcg}|j}d\}}tdD]}|dk(r|n |d|dz }|d}|jd\} } |dk(rMtj|dz|dz|j d fd tj } | | } } |||| z|| zf}n|dk(r ||| z || z|f}n|d k(r||z|| z ||z| z|f}n|dk(r| z||| z| z|| zf}n|d k(r| z||z|| z| z||z| zf}nw|d k(r| z| z | z|| z|| z| zf}nY|dk(r| z|z | z | z|| z|z || z| zf}n5|dk(r|| z | z| z ||| zf}n|dk(r|| z | z|z | z ||| z|z f}dd \}}d|D\}}}}|||z d||z df ||||f<| | }}|j|||jdz||jdz}|j||j|} |jd |jd|jd |jdf|d<|S)zCreate a 3x3 image mosaic.)rr\rrCr rwrxr|r]ryrzr[Nc34K|]}t|dyw)rN)rrsrts rrvz"Mosaic._mosaic9..s3Ac!Qis) r_rhrLr}r~rrrr`r0r)rrrruhpwprNrrwrrimg9h0w0crrx1y1x2y2rs rrpzMosaic._mosaic9s JJBqA%&!V6 1Ea!e1LLu%C##O4DAqAvwwAq1uciil;SQABq!a%Q&aq1ua!eQ&aFAE1r6A:q0aFAq2vz1q50aFAFAFQJB :aFQJBBB :aFRK!OQVQVb[!b&1*DaE1r6A:q!b&0aE1r6B;?Aq2v{:2AJD$33NBB"%R$YZd%;!#ADKKPQN?SWS^S^_`SaCa#ab Urc|djdd\}}|djd|dj|||dj|||S)Update labels.rwNr] instancesxyxyformat)r convert_bbox denormalize add_padding)rrrnhnws rrzMosaic._update_labelssc$$Ra(B{(((7{''B/{''d3 rct|dk(riSg}g}|jdz}|D]*}|j|d|j|d,|dd|dd||ftj|dt j|d|j d}|dj|||dj}|d||d<|S) z3Return labels with mosaic border instances clipped.rr]clsrim_file ori_shapeaxis)rrrxrr mosaic_border) rjr_r0r} concatenaterr`clipremove_zero_area_boxes)rrrrr_rrgoods rrzMosaic._cat_labelss }  "I  Q#F JJve} %   VK0 1$%Q' 2&q)+6#U^>>#q)"..yqA![[ *  [!&&ue4K(??A*51$7 Ur)rYr[)T) r#r$r%r&rrFrKrorp staticmethodrr __classcell__r8s@rrWrWs> YO "H.`rrWc2eZdZdZddfd ZdZdZxZS)MixUpz5Class for applying MixUp augmentation to the dataset.c*t||||y)zXInitializes MixUp object with dataset, pre_transform, and probability of applying MixUp.r>N)r^r)rr?r@rAr8s rrzMixUp.__init__s  KrcZtjdt|jdz S)z$Get a random index from the dataset.rr )rDrirjr?rs rrFzMixUp.get_indexes#s!~~aT\\!2Q!677rcDtjjdd}|dd}|d|z|dd|z zzjtj|d<t j |d|dgd|d<tj |d|dgd|d<|S) zGApplies MixUp augmentation as per https://arxiv.org/pdf/1710.09412.pdf.g@@rCrrwr rrr)r}rDbetaastyperrr)rrrlabels2s rrKzMixUp._mix_transform's IINN4 &&q)*WU^q1u-EEMMbhhWu '33VK5H'R]J^4_fgh{u wu~'FJu  rrTr!)r#r$r%r&rrFrKrrs@rrrs?L8rrcLeZdZdZ d dZdZdZdZdZdZ d d Z y) RandomPerspectivea Implements random perspective and affine transformations on images and corresponding bounding boxes, segments, and keypoints. These transformations include rotation, translation, scaling, and shearing. The class also offers the option to apply these transformations conditionally with a specified probability. Attributes: degrees (float): Degree range for random rotations. translate (float): Fraction of total width and height for random translation. scale (float): Scaling factor interval, e.g., a scale factor of 0.1 allows a resize between 90%-110%. shear (float): Shear intensity (angle in degrees). perspective (float): Perspective distortion factor. border (tuple): Tuple specifying mosaic border. pre_transform (callable): A function/transform to apply to the image before starting the random transformation. Methods: affine_transform(img, border): Applies a series of affine transformations to the image. apply_bboxes(bboxes, M): Transforms bounding boxes using the calculated affine matrix. apply_segments(segments, M): Transforms segments and generates new bounding boxes. apply_keypoints(keypoints, M): Transforms keypoints. __call__(labels): Main method to apply transformations to both images and their corresponding annotations. box_candidates(box1, box2): Filters out bounding boxes that don't meet certain criteria post-transformation. Ncf||_||_||_||_||_||_||_y)zDInitializes RandomPerspective object with transformation parameters.N)degrees translatescaleshear perspectiver`r@)rrrrrrr`r@s rrzRandomPerspective.__init__Is7 "  & *rctjdtj}|jd dz |d<|jd dz |d<tjdtj}t j |j |j |d<t j |j |j |d <tjdtj}t j |j |j}t j d|jz d|jz}tj|d | |d dtjdtj}tjt j |j |jtjzd z |d<tjt j |j |jtjzd z |d<tjdtj} t j d|jz d|jz|j dz| d<t j d|jz d|jz|j dz| d<| |z|z|z|z} |ddk7s.|ddk7s&| tjdk7j#rV|j r$tj$|| |j d}n&tj&|| d d|j d}|| |fS)a[ Applies a sequence of affine transformations centered around the image center. Args: img (ndarray): Input image. border (tuple): Border dimensions. Returns: img (ndarray): Transformed image. M (ndarray): Transformation matrix. s (float): Scale factor. r|rzr r])rr]r)r r])r]r)r]r rr)anglecenterrN)rr )r r?ryryry)dsize borderValue)r}eyefloat32rrDrErrrcv2getRotationMatrix2DmathtanrpirsizeanywarpPerspective warpAffine) rrwr`CPRaruSTMs raffine_transformz"RandomPerspective.affine_transform[s FF1BJJ '99Q<-!#$99Q<-!#$ FF1BJJ '..$"2"2!2D4D4DE$..$"2"2!2D4D4DE$ FF1BJJ ' NNDLL=$,, 7 NN1tzz>1tzz> :''aaH"1 FF1BJJ '((6>>4::+tzzBTWWLsRS$((6>>4::+tzzBTWWLsRS$ FF1BJJ '..t~~!5sT^^7KLtyyYZ|[$..t~~!5sT^^7KLtyyYZ|[$ EAIMA  1INq QAN3G3G3I))#q _nnS!BQ%tyyo^Aqyrc|t|}|dk(r|Stj|dzdf|j}|ddgdfj |dzd|ddddf<||j z}|j r|ddddf|ddddfz n |ddddfj |d}|ddgd f}|ddgd f}tj|jd |jd |jd |jd f|jj d|j S) a Apply affine to bboxes only. Args: bboxes (ndarray): list of bboxes, xyxy format, with shape (num_bboxes, 4). M (ndarray): affine matrix. Returns: new_bboxes (ndarray): bboxes after affine, [num_bboxes, 4]. rr[r|rzN)rr r]r|rr|r]r r]r)rr]r[r)r r|rrr ) rjr}onesr{reshaperrrrr)rbboxesrraxyrtys r apply_bboxeszRandomPerspective.apply_bboxess' K 6M WWa!eQZv|| 41667??AqI1bqb5 !##X(,(8(8bBQBi"Q!V*$bBQBi P PQRTU V q,  q, ~~quuQxq1558QUU1XFfll[ccdeghikkkrc |jdd\}}|dk(rg|fStj||zdf|j}|j dd}||ddddf<||j z}|ddddf|ddddfz }|j |dd}tj |Dcgc])}t||jd|jd+c}d}||fScc}w)a| Apply affine to segments and generate new bboxes from segments. Args: segments (ndarray): list of segments, [num_samples, 500, 2]. M (ndarray): affine matrix. Returns: new_segments (ndarray): list of segments after affine, [num_samples, 500, 2]. new_bboxes (ndarray): bboxes after affine, [N, 4]. Nr]rr|rzrr ) rr}rr{rrstackr r)rsegmentsrranumrrs rapply_segmentsz RandomPerspective.apply_segmentss#3 6x<  WWa#gq\ 8##B*1bqb5 !##X 2A2YAqsF #::aQ'QYZQY2;r499Q<1FQYZ\]^x[s+.C%c>|jdd\}}|dk(r|Stj||zdf|j}|dj ||zd}|dddfj ||zd|ddddf<||j z}|ddddf|ddddfz }|dddfdk|dddfdkz|dddf|j dkDz|dddf|j dkDz}d||<tj||gd j ||dS) z Apply affine to keypoints. Args: keypoints (ndarray): keypoints, [N, 17, 3]. M (ndarray): affine matrix. Returns: new_keypoints (ndarray): keypoints after affine, [N, 17, 3]. Nr]rr|rz).r]r .rr)rr}rr{rrrr)r keypointsrrankptrvisibleout_masks rapply_keypointsz!RandomPerspective.apply_keypointssH//"1%4 6  WWa$h])// :F#++AHa8c2A2g&..q4x;1bqb5 !##X 2A2YAqsF #q!tHqLR1X\2bAh16MNRTUVXYUYRZ]a]f]fgh]iRij~~r7m"5==aqIIrc|jrd|vr|j|}|jdd|d}|d}|jd}|jd|j|jdd ddd |jd|j }|jd |d d zz|jd |d d zzf|_|j||\}}}|j|j|}|j} |j} t| r|j| |\}} | |j| |} t|| | dd } | j |j |j#||d|j%|jj&| jj&t| rdnd} | | |d<|| |d<||d<|jdd |d<|S)z Affine images and targets. Args: labels (dict): a dict of `bboxes`, `segments`, `keypoints`. r ratio_padNrwrrrrr]rr rF) bbox_format normalizedT)scale_wscale_h bbox_only{Gz?皙?)box1box2area_thrrx)r@rLrrrr`rrrrrrrjrrrrrbox_candidatesr) rrrwrrr`rrrrr new_instancesrNs rr zRandomPerspective.__call__s   /"?''/F ;%UmUmJJ{+ f- syy!}TrT23OT[[9IIaL6!9q=0#))A,Q2NN --c6: Q""9#3#3Q7%%'' x=#228Q? FH  ,,Y:I!&(I6^cd  DII& uE   Y%5%5%7%7%2%9%9%;%;14XD  J,A.{Au u "%))BQ- rc|d|dz |d|dz }}|d|dz |d|dz } } tj| | |zz | | |zz } | |kD| |kDz| | z||z|zz |kDz| |kzS)a Compute box candidates based on a set of thresholds. This method compares the characteristics of the boxes before and after augmentation to decide whether a box is a candidate for further processing. Args: box1 (numpy.ndarray): The 4,n bounding box before augmentation, represented as [x1, y1, x2, y2]. box2 (numpy.ndarray): The 4,n bounding box after augmentation, represented as [x1, y1, x2, y2]. wh_thr (float, optional): The width and height threshold in pixels. Default is 2. ar_thr (float, optional): The aspect ratio threshold. Default is 100. area_thr (float, optional): The area ratio threshold. Default is 0.1. eps (float, optional): A small epsilon value to prevent division by zero. Default is 1e-16. Returns: (numpy.ndarray): A boolean array indicating which boxes are candidates based on the given thresholds. r]rr|r )r}maximum) rr r wh_thrar_thrr epsw1h1w2h2ars rrz RandomPerspective.box_candidates s a47"DGd1g$5Ba47"DGd1g$5B ZZb3hrCx 9V V ,R27S=0IH0TUY[^dYdeer)rUr rrUrUrN)r]dr gؗҜ<) r#r$r%r&rrrrrr rrrrrr1sF0 #+$3jl2 2J.0dfrrc eZdZdZdddZdZy) RandomHSVz This class is responsible for performing random adjustments to the Hue, Saturation, and Value (HSV) channels of an image. The adjustments are random but within limits set by hgain, sgain, and vgain. Nc.||_||_||_y)aU Initialize RandomHSV class with gains for each HSV channel. Args: hgain (float, optional): Maximum variation for hue. Default is 0.5. sgain (float, optional): Maximum variation for saturation. Default is 0.5. vgain (float, optional): Maximum variation for value. Default is 0.5. Nhgainsgainvgain)rrr r!s rrzRandomHSV.__init__)s   rc|d}|js|js |jrtjj ddd|j|j|jgzdz}t jt j|t j\}}}|j}tjdd|j}||dzdzj|} tj||dzdd j|} tj||d zdd j|} t jt j|| t j|| t j|| f} t j| t j | |S) z Applies random HSV augmentation to an image within the predefined limits. The modified image replaces the original image in the input 'labels' dict. rwrr r|rrzrr])dst)rr r!r}rDrErsplitcvtColor COLOR_BGR2HSVr{arangerrmergeLUT COLOR_HSV2BGR) rrrwrhuesatvalr{rtlut_huelut_satlut_valim_hsvs rr zRandomHSV.__call__6sT Um ::tzz !!"a+tzz4::tzz.RRUVVAIIcll38I8I&JKMCcIIE !S0AAaDC'//6Ggga!A$h3/66u=Ggga!A$h3/66u=GYYW 5swwsG7LcggVY[bNcdeF LL!2!2 < r)rrrr!r#r$r%r&rr rrrrr!s rrc eZdZdZdddZdZy) RandomFlipz Applies a random horizontal or vertical flip to an image with a given probability. Also updates any instances (bounding boxes, keypoints, etc.) accordingly. Ncj|dvs Jd|d|cxkrdksJJ||_||_||_y)a Initializes the RandomFlip class with probability and direction. Args: p (float, optional): The probability of applying the flip. Must be between 0 and 1. Default is 0.5. direction (str, optional): The direction to apply the flip. Must be 'horizontal' or 'vertical'. Default is 'horizontal'. flip_idx (array-like, optional): Index mapping for flipping keypoints, if any. ) horizontalverticalz2Support direction `horizontal` or `vertical`, got rrYN)rA directionflip_idx)rrAr:r;s rrzRandomFlip.__init__SsL66x:lmvlw8xx6A}}}}"  rc|d}|jd}|jd|jdd\}}|jrdn|}|jrdn|}|jdk(rGt j |j kr&tj|}|j||jd k(rt j |j krvtj|}|j||jD|j8tj|jdd|jddf|_ tj||d<||d<|S) a Applies random flip to an image and updates any instances like bounding boxes or keypoints accordingly. Args: labels (dict): A dictionary containing the keys 'img' and 'instances'. 'img' is the image to be flipped. 'instances' is an object containing bounding boxes and optionally keypoints. Returns: (dict): The same dict with the flipped image and updated instances under the 'img' and 'instances' keys. rwrxywhrNr]r r9r8) rLrrrr:rDrAr}flipudfliplrr;rascontiguousarray)rrrwrrrs rr zRandomFlip.__call__ds4UmJJ{+ f-yy!}1%%A1%%A1 >>Z 'FMMOdff,D))C.C   Q  >>\ )fmmo.F))C.C   Q }}(Y-@-@-L&(&:&:9;N;NqRVR_R_abOb;c&d #,,S1u '{ r)rr8Nr!r4rrrr6r6Ls !"rr6c&eZdZdZddZddZdZy) LetterBoxzDResize image and padding for detection, instance segmentation, pose.cX||_||_||_||_||_||_y)z5Initialize LetterBox object with specific parameters.N) new_shapeauto scaleFillscaleupstrider)rrDrErFrGrrHs rrzLetterBox.__init__s," "   rNc |i}||jdn|}|jdd}|jd|j}t |t r||f}t |d|dz |d|dz }|js t |d}||f}t t|d|zt t|d|zf}|d|dz |d|dz } } |jrAtj| |jtj| |j} } n1|jr%d\} } |d|df}|d|dz |d|dz f}|jr | dz} | dz} |ddd |k7r&tj ||tj" }|jrt t| d z ndt t| d z} } |jrt t| d z ndt t| d z}} tj$|| | | |tj&d }|jdr |d| | ff|d<t)|r |j+||| | }||d<||d<|S|S)z2Return updated labels and image with added border.Nrwr]rmrr rY)rUrUr interpolationr r)valuerrx)rnrrLrDrGrHrrGroundrEr}modrHrFrrresize INTER_LINEARcopyMakeBorderBORDER_CONSTANTrjr)rrimagerwrrDrratio new_unpaddwdhtopbottomleftrights rr zLetterBox.__call__s} >F#(=fjje "1 JJ|T^^< i %"I.I  ! uQx'1a)@ A||As A1eAhl+,c%a1 2E.FF 1 ! ,ilYq\.IB 99VVB ,bffR.EB ^^FB"1y|4IaL58+Yq\E!H-DDE ;; !GB !GB 2;) #**S)3;K;KLC.2kkc%S/*q#eBQTHoBVV.2kkc%S/*q#eBQTHoBVe  c64@S@S'68 ::k "#)+#6s "DF;  v;((B?FF5M&/F? #MJrc|djd|dj|djddddd|dj||dj |||S)rrrrrwNr]r)rrrrr)rrrTrrs rrzLetterBox._update_labelssv{(((7'{'')<).(s,*h_`Z[dedgdgaS>>:PRT+U_`,,)rAr1ralbumentationsr __version__Blur MedianBlurToGrayCLAHERandomBrightnessContrast RandomGammaImageCompressionr( BboxParamsrinfor9 ImportError Exception)rrAprefixAres rrzAlbumentations.__init__s,- ( & !--t < t $4 $**S*1  $""s";=AYYqall&`n_ol6pYqDN KK*h_`*h!hh i   ( KK6(1# ' ' (sC.D EED;;Ec$|d}|d}t|r|djd|dj|jddddd|dj}|j rt j |jkrp|j |||}t|d d kDrK|d |d<tj|d |d<tj|d tj }|dj||S)zLGenerates object detections and returns a dictionary with detection results.rwrrr=Nr]r)rSrr}r}rrSrrz)r) rjr normalizerrr1rDrAr}arrayrupdate)rrrmrrnews rr zAlbumentations.__call__.s E]Um s8 ;  , ,V 4 )F;  ) )288BQ<"+= >K(//F~~&--/DFF":nn2f3nOs>*+a/$'LF5M$&HHS-@$AF5MXXc(m2::FF ;  & &f & 5 rN)rYr4rrrrtrt s(4rrtc8eZdZdZ ddZdZdZdZy)Formata Formats image annotations for object detection, instance segmentation, and pose estimation tasks. The class standardizes the image and instance annotations to be used by the `collate_fn` in PyTorch DataLoader. Attributes: bbox_format (str): Format for bounding boxes. Default is 'xywh'. normalize (bool): Whether to normalize bounding boxes. Default is True. return_mask (bool): Return instance masks for segmentation. Default is False. return_keypoint (bool): Return keypoints for pose estimation. Default is False. mask_ratio (int): Downsample ratio for masks. Default is 4. mask_overlap (bool): Whether to overlap masks. Default is True. batch_idx (bool): Keep batch indexes. Default is True. cf||_||_||_||_||_||_||_y)z3Initializes the Format class with given parameters.N)rr return_maskreturn_keypoint mask_ratio mask_overlap batch_idx)rrrrrrrrs rrzFormat.__init__Qs9'"&.$("rc|jd}|jdd\}}|jd}|jd}|j|j|j ||t |}|j r|r.|j||||\}}}tj|}nYtj|jrdn||jd|jz|jd|jz}||d <|jr|j|||j||d<|rtj|ntj||d<|rtj|jntj|d f|d <|j r"tj|j"|d <|j$rtj||d <|S)zWReturn formatted image, classes, bounding boxes & keypoints to be used by 'collate_fn'.rwNr]rrrr rmasksr[rrr)rLrrrrrjr_format_segmentstorch from_numpyrdrrr _format_imgrrrr) rrrwrrrrnlrs rr zFormat.__call__bsjjyy!}1jjJJ{+ d&6&67a# ^   (,(=(=iaQR(S%y#((/ ):):ACIIaLTXTcTcDc$'IIaLDOO$CE#F7O >>   1 %((-u 13((-Ru AC5++I,<,<=VXZ[U\I]x   "'"2"293F3F"GF;  >>"'++b/F;  rct|jdkrtj|d}tj|j dddddd}t j|}|S)z=Format the image for YOLO from Numpy array to PyTorch tensor.r|rr]rr N)rjrr} expand_dimsr@ transposerr)rrws rrzFormat._format_imgs] syy>A ..b)C""3==Aq#9$B$#?@s# rc|j}|jr-t||f||j\}}|d}||}||}nt ||f|d|j}|||fS)z!Convert polygon points to bitmap.)downsample_ratioNr )colorr)rrr rr )rrrrrrr sorted_idxs rrzFormat._format_segmentssu%%    61vxZ^ZiZi j E:$KE!*-Ij/C"Aq681t_Ei$$rN)r=TFFr[TT)r#r$r%r&rr rrrrrrrBs2 $"!&"#"< %rrcjtt|||jt|jt |j |j|j|j|j|rdn t||fg}|jjdg}|jr}|jjdd}t|dk(r,|j d kDrd |_t#j$d n'|r%t||dk7rt'd |d |dt|t)|||j* t-dt/|j0|j2|j4t7d|j8t7d|j |gS)z6Convert images to a size suitable for YOLOv8 training.)r_rArbN)rD)rrrrrr@r; kpt_shaperrUuZWARNING ⚠️ No 'flip_idx' array defined in data.yaml, setting augmentation 'fliplr=0.0'zdata.yaml flip_idx=z& length must be equal to kpt_shape[0]=)r@rArYrr9)r:rAr8)r:rAr;)r(rWmosaicr` copy_pasterrrrrrrBr-rn use_keypointsrjr?rwarning ValueErrorrmixuprtrhsv_hhsv_shsv_vr6r>)r?r_hypstretchr@r;rs r v8_transformsrsdweszz2CNN#KKmm))))")$yE5>/R     M|| B/HLL$$[$7 x=A #**s"2CJ NNw x 3x=IaL828*? 99Z  rc||||fS)zgMap HSV (hue, saturation, value) jitter into ColorJitter values (brightness, contrast, saturation, hue)r)rruvs rhsv2colorjitterrs aA:rc td} ddl} ddlm} t | j dd|r| j |||g}| rtj| d n|dkDr|| j| gz }|dkDr|| j| gz }t|||frD|| jt|||gz }n%| j| | j|| g}|| j!|| | gz }tj| dj#d|Dz| j%|S#t&$rYyt($r#}tj| |Yd}~yd}~wwxYw)zSYOLOv8 classification Albumentations (optional, only used if package is installed).rvrN) ToTensorV2rwTrx)heightwidthrz.auto augmentations are currently not supportedrb)max_size)rr)rrr6c3\K|]$}|js|jdd&ywrrrs rrvz*classify_albumentations..s+&d[\VW`a`c`c!~~6Lb'Q[\r)rralbumentations.pytorchrrrRandomResizedCroprrHorizontalFlip VerticalFlipr ColorJitterrSmallestMaxSizerrr9r(rr)augmentrrhflipvfliprrrrrauto_augrrrrrs rclassify_albumentationsrst( )F$"5ammW48 $$DE$JKA vh&TUV19!**U*344A19!..5.122Aue,--!--u)MNOOA""D"11<$D  rc |jdd\}}tj|z j|z }t ||zt ||z}}j r fd||fDnjjf\}}t ||z dz dz t ||z dz dz } } t j||dfd|j} tj|||ftj| | | |z| | |zf<| S) a  Resizes the image and pads it with a letterbox method. Args: im (numpy.ndarray): The input image as a numpy array of shape HWC. Returns: (numpy.ndarray): The letterboxed and resized image as a numpy array. Nr]c3|K|]3}tj|jz jz5yw)N)rceilrH)rsrtrs rrvz-ClassifyLetterBox.__call__..s,KFq$))A O,t{{:Fs9<r r|ryrzrJ) rrrrrMrEr}r~r{rrOrP) rrmimhimwrrrhswsrXrZim_outs ` rr zClassifyLetterBox.__call__ s88BQ<S  dffsl +S1W~uS1W~1PTyyKQFK_c_e_egkgmgm^nB26Q,,-ub1f\C5G/HT"b!c:-0ZZQFRURbRb-cs37{DM)* r)r]Fr^r#r$r%r&rr rrs@rrrs  rrc*eZdZdZdfd ZdZxZS)rzYOLOv8 CenterCrop class for image preprocessing, designed to be part of a transformation pipeline, e.g., T.Compose([CenterCrop(size), ToTensor()]). cht|t|tr||fn|\|_|_y)z5Converts an image from numpy array to PyTorch tensor.N)r^rrGrHrr)rrr8s rrzCenterCrop.__init__(s* )3D#)>$Drc|jdd\}}t||}||z dz||z dz}}tj||||z|||zf|j|j ftj S)a Resizes and crops the center of the image using a letterbox method. Args: im (numpy.ndarray): The input image as a numpy array of shape HWC. Returns: (numpy.ndarray): The center-cropped and resized image as a numpy array. Nr]rJ)rrrrOrrrP)rrmrrmrXrZs rr zCenterCrop.__call__-s|88BQ<S SM1WNS1WNTzz"Sq[$tax-78466466:JZ]ZjZjkkr)rrrs@rrr#sI lrrc*eZdZdZdfd ZdZxZS)rz^YOLOv8 ToTensor class for image preprocessing, i.e., T.Compose([LetterBox(size), ToTensor()]).c0t|||_y)zGInitialize YOLOv8 ToTensor object with optional half-precision support.N)r^rhalf)rrr8s rrzToTensor.__init__@s  rctj|jdddd}tj|}|j r|j n|j }|dz}|S)au Transforms an image from a numpy array to a PyTorch tensor, applying optional half-precision and normalization. Args: im (numpy.ndarray): Input image as a numpy array with shape (H, W, C) in BGR order. Returns: (torch.Tensor): The transformed image as a PyTorch tensor in float32 or float16, normalized to [0, 1]. )r]rr Nrgo@)r}r@rrrrfloat)rrms rr zToTensor.__call__Es] ! !",,y"9$B$"? @   b !))RWWY e  rFrrs@rrr=sh rrr)FrUrUrUrYrYrY) Tr)g{Gz?rYrrUgQ?gffffff?g?rrF),rrDcopyrrnumpyr}rtorchvision.transformsr+rultralytics.utilsrrultralytics.utils.checksrultralytics.utils.instancerultralytics.utils.metricsrultralytics.utils.opsr utilsr r rr(r<rWrrrr6rBr`rtrrrrrrrrrrrrs   ".20.-9#$#$L\\4*"*"ZV Vr *mfmf`((V66rAAH??D33nQ%Q%hN@!    )$X//dll4r