/j dZddlmZddlZddlZddlmZddlmZddl m Z ddl Z ddl Z ddlmZmZmZmZmZmZe jgde j d z Ze jgd ZdbdcdZdddedZ dfdgdZdddhd!Z dddid'Zdjd*Zdbdkd-Zdddld/Z dmdnd2Z!Gd3d4eZ"dodpd8Z#eed9idfdqdAZ$eedBidCdDdfdrdHZ%dsdLZ&d deidMdNfdtdVZ'GdWdDeZ(GdXdYeeZ)GdZd[e)Z*Gd\d]e)Z+Gd^d_eeZ,Gd`dae)Z-dS)uzModel validation metrics.) annotationsN) defaultdict)Path)Any)LOGGERDataExportMixin SimpleClass TryExceptchecks plt_settings)gp= ף??r ffffff?rHzG?r ףp= ?rףp= ?rQ?rףp= ?r{Gz?r)dtypeg$@)?rrrrrr333333?r?rrrrrrrFHz>box1 np.ndarraybox2iouboolepsfloatreturnc|j\}}}}|j\}} } } tj|dddf| tj|dddf|z dtj|dddf| tj|dddf| z dz} | |z | | z z} |r||z ||z z}| |dddfz| z } | | |zz S)a?Calculate the intersection over box2 area given box1 and box2. Args: box1 (np.ndarray): A numpy array of shape (N, 4) representing N bounding boxes in x1y1x2y2 format. box2 (np.ndarray): A numpy array of shape (M, 4) representing M bounding boxes in x1y1x2y2 format. iou (bool, optional): Calculate the standard IoU if True else return inter_area/box2_area. eps (float, optional): A small value to avoid division by zero. Returns: (np.ndarray): A numpy array of shape (N, M) representing the intersection over box2 area. Nr)Tnpminimummaximumclip)rrrrb1_x1b1_y1b1_x2b1_y2b2_x1b2_y1b2_x2b2_y2 inter_areaarea box1_areas ^/home/longshao/multi-rider-rag/.venv/lib/python3.11/site-packages/ultralytics/utils/metrics.pybbox_ioar4s"&E5%!%E5%*U111d7^U33bjqqq$wQV6W6WW]]^_`` 5D>5))BJuQQQW~u,M,MM d1ggJ EMeem ,D 6U]uu}5 i4((:5  $$ torch.Tensorc|ddd|dddc\}}\}}tj||tj||z dd}|||z d||z dz|z |zz S)aBCalculate intersection-over-union (IoU) of boxes. Args: box1 (torch.Tensor): A tensor of shape (N, 4) representing N bounding boxes in (x1, y1, x2, y2) format. box2 (torch.Tensor): A tensor of shape (M, 4) representing M bounding boxes in (x1, y1, x2, y2) format. eps (float, optional): A small value to avoid division by zero. Returns: (torch.Tensor): An NxM tensor containing the pairwise IoU values for every element in box1 and box2. References: https://github.com/pytorch/vision/blob/main/torchvision/ops/boxes.py r)r unsqueezechunktorchminmaxclamp_prod)rrra1a2b1b2inters r3box_iourF:s //2288A>> @V@VWX@Y@Y@_@_`acd@e@eHRhr2 Yr2  2r!2!2 2 : :1 = = B B1 E EE R"WNN1%%bq(9(99EACG HHr5TxywhGIoUDIoUCIoUc|r}|dd|ddc\}}} } \} } } }| dz | dz | dz |dz f\}}}}||z ||z||z ||zf\}}}}| |z | |z| |z | |zf\}}}}nP|dd\}}}}|dd\}}}}||z ||z |z} } ||z ||z |z}} ||||z d||||z dz}| | z| |zz|z |z}||z }|s|s|r||||z }||||z }|s|r|d|dz|z} ||z|z |z d||z|z |z dzdz }!|rdt jdzz | |z | | z z dz}"tj 5|"|"|z d|zzz }#dddn #1swxYwY||!| z |"|#zzz S||!| z z S||z|z}$||$|z |$z z S|S)axCalculate the Intersection over Union (IoU) between bounding boxes. This function supports various shapes for `box1` and `box2` as long as the last dimension is 4. For instance, you may pass tensors shaped like (4,), (N, 4), (B, N, 4), or (B, N, 1, 4). Internally, the code will split the last dimension into (x, y, w, h) if `xywh=True`, or (x1, y1, x2, y2) if `xywh=False`. Args: box1 (torch.Tensor): A tensor representing one or more bounding boxes, with the last dimension being 4. box2 (torch.Tensor): A tensor representing one or more bounding boxes, with the last dimension being 4. xywh (bool, optional): If True, input boxes are in (x, y, w, h) format. If False, input boxes are in (x1, y1, x2, y2) format. GIoU (bool, optional): If True, calculate Generalized IoU. DIoU (bool, optional): If True, calculate Distance IoU. CIoU (bool, optional): If True, calculate Complete IoU. eps (float, optional): A small value to avoid division by zero. Returns: (torch.Tensor): IoU, GIoU, DIoU, or CIoU values depending on the specified flags. r9rr8N) r;r%r&r?powmathpiatanr<no_grad)%rrrGrHrIrJrx1y1w1h1x2y2w2h2w1_h1_w2_h2_r(r*r)r+r,r.r-r/rEunionrcwchc2rho2valphac_areas% r3bbox_iourgQs: 4-1ZZ2->-> 1b@Q@Q*RR*2r2r!VR!VR!VR!V;S#s%'#XrCxc28%K"ueU%'#XrCxc28%K"ueUU%)ZZ2%6%6"ueU%)ZZ2%6%6"ueU  3B  3B]]5 ! !EMM%$8$8 8 @ @ C C eu}}U333 fQii E Gb2g  % +E %-C /t/t/ ]]5 ! !EMM%$8$8 8 ]]5 ! !EMM%$8$8 8  #4 #RVVAYY&,B&.33A66%%-%:ORW:W9\9\]^9_9__D 5!^b(8(8BG>>;K;K(K'P'PQR'S'SS]__66SAG!45E666666666666666dRi!e)344? "b3fun... JsJ88J<?J<mask1mask2ctj||jd}|ddddf|ddz|z }|||zz S)aCalculate masks IoU. Args: mask1 (torch.Tensor): A tensor of shape (N, n) where N is the number of ground truth objects and n is the product of image width and height. mask2 (torch.Tensor): A tensor of shape (M, n) where M is the number of predicted objects and n is the product of image width and height. eps (float, optional): A small value to avoid division by zero. Returns: (torch.Tensor): A tensor of shape (N, M) representing masks IoU. rr8N)r<matmulr#r?sum)rhrir intersectionr_s r3mask_iournsi<uw//66q99L YYq\\!!!T' "UYYq\\$%7 7< GE 53; ''r5kpt1kpt2r1sigma list[float]c|ddddddf|dz d|ddddddf|dz dz}tj||j|j}|ddk}|d|zd|ddddf|zzdzz }| |dddfzd |d dddf|zz S) aRCalculate Object Keypoint Similarity (OKS). Args: kpt1 (torch.Tensor): A tensor of shape (N, 17, 3) representing ground truth keypoints. kpt2 (torch.Tensor): A tensor of shape (M, 17, 3) representing predicted keypoints. area (torch.Tensor): A tensor of shape (N,) representing areas from ground truth. sigma (list[float]): A list containing 17 values representing keypoint scales. eps (float, optional): A small value to avoid division by zero. Returns: (torch.Tensor): A tensor of shape (N, M) representing keypoint similarities. Nr).rr9r8).r8)devicer).r9rM)rNr<tensorrtrexprl)rorpr1rqrdkpt_maskes r3kpt_iourzs aaaqqq!m tF| +0033tAAAtQQQM7JTRX\7Y6^6^_`6a6aaA Lt{$* E E EEF|q H a%i__Q  44 #6#< = ABARHHJJ!!!T'* * / / 3 3x||B7G7G47PSV7V WWr5boxes/tuple[torch.Tensor, torch.Tensor, torch.Tensor]ctj|ddddfddz |ddddffd}|dd\}}}|}|}|d}|d}||z||zz||z||zz||z |z|zfS)azGenerate covariance matrix from oriented bounding boxes. Args: boxes (torch.Tensor): A tensor of shape (N, 5) representing rotated bounding boxes, with xywhr format. Returns: (tuple[torch.Tensor, torch.Tensor, torch.Tensor]): Covariance matrix components (a, b, c) where the covariance matrix is [[a, c], [c, b]], each of shape (N, 1). Nr9rL rMdimr8)r<catrNsplitcossin) r{gbbsabcrrcos2sin2s r3_get_covariance_matrixrs 9eAAAqsFm''**R/qqq!""u>B G G GDjjj##GAq! %%''C %%''C 771::D 771::D t8a$h D1t8 3a!es]S5H HHr5obb1obb2c|dddfdd\}}|dddfdd\}}t|\}} } t|\} } } || z||z dz| | z||z dzz|| z| | zz| | zdz |zz dz}| | z||z z||z z|| z| | zz| | zdz |zz dz}|| z| | zz| | zdz d || z| dz d | | z| dz d zz|zz |zdz}||z|z|d }d | z |z}d|z }|r|ddd fdd\}}|ddd fdd\}}d tj dzz ||z ||z z dz}tj 5|||z d|zzz }dddn #1swxYwY|||zz S|S) a8Calculate probabilistic IoU between oriented bounding boxes. Args: obb1 (torch.Tensor): Ground truth OBBs, shape (N, 5), format xywhr. obb2 (torch.Tensor): Predicted OBBs, shape (N, 5), format xywhr. CIoU (bool, optional): If True, calculate CIoU. eps (float, optional): Small value to avoid division by zero. Returns: (torch.Tensor): OBB similarities, shape (N,). Notes: OBB format: [center_x, center_y, width, height, rotation_angle]. References: https://arxiv.org/pdf/2106.06072v1.pdf .Nr9r8rMrr ?rLrY@r) rrrNr?sqrtlogclamprvrOrPrQr<rR)rrrJrrSrTrWrXrArCc1rBrDrbt1t2t3bdhdrrUrVrYrZrdres r3probioursb$#rr']  + +FB #rr']  + +FB'--JBB'--JBB r'b2g]]1%% %bR"WMM!4D4D(D D"r'VX[]V]I^bdgibiananopaqaqIqtwIwx  B GR BG ,"r'b2g1F"r'WXIYIY1Y\_1_ `dg gB r'b2g "r'q!1!1 1 b266!99$,,Q//27RVVAYY3F2N2Nq2Q2QQWWYY Y\_ _ a   ceec  B r'B,  c5 ) )B " c ! ' ' ) )B b&C c1Q3h%%aR%00Bc1Q3h%%aR%00B !^b 0 0BG>>3C3C CHHKK K ]__ . .SAG,-E . . . . . . . . . . . . . . .QY JsK  K$'K$torch.Tensor | np.ndarrayct|tjrtj|n|}t|tjrtj|n|}|dddfdd\}}d|dddfddD\}}t |\}}} dt |D\} } } || z||z dz|| z||z dzz|| z|| zz| | zdz |zz d z} | | z||z z||z z|| z|| zz| | zdz |zz d z}|| z|| zz| | zdz d ||z| dz d | | z| dz d z z|zz |z d z}| |z|z |d }d| z |z }d|z S)aCalculate the probabilistic IoU between oriented bounding boxes. Args: obb1 (torch.Tensor | np.ndarray): A tensor of shape (N, 5) representing ground truth obbs, with xywhr format. obb2 (torch.Tensor | np.ndarray): A tensor of shape (M, 5) representing predicted obbs, with xywhr format. eps (float, optional): A small value to avoid division by zero. Returns: (torch.Tensor): A tensor of shape (N, M) representing obb similarities. References: https://arxiv.org/pdf/2106.06072v1.pdf .Nr9r8rMrc3LK|]}|ddV dSrMNsqueeze.0xs r3 z batch_probiou..s1 J JaaiimmD! J J J J J Jr5c3LK|]}|ddV dSrrrs r3rz batch_probiou..s1LL!!))B--%LLLLLLr5r rrLrrr) isinstancer$ndarrayr< from_numpyrrrNr?rrrrv)rrrrSrTrWrXrArCrrBrDrbrrrrrs r3 batch_probiours&0bj%A%A K5 D ! ! !tD%/bj%A%A K5 D ! ! !tD #rr']  + +FB J J4RaR=+>+>qb+>+I+I J J JFB'--JBBLL/Ed/K/KLLLJBB r'b2g]]1%% %bR"WMM!4D4D(D D"r'VX[]V]I^bdgibiananopaqaqIqtwIwx  B GR BG ,"r'b2g1F"r'WXIYIY1Y\_1_ `dg gB r'b2g "r'q!1!1 1 b266!99$,,Q//27RVVAYY3F2N2Nq2Q2QQWWYY Y\_ _ a   ceec  B r'B,  c5 ) )B " c ! ' ' ) )B r6Mr5皙?tuple[float, float]cdd|zz d|zfS)a}Compute smoothed positive and negative Binary Cross-Entropy targets. Args: eps (float, optional): The epsilon value for label smoothing. Returns: pos (float): Positive label smoothing BCE target. neg (float): Negative label smoothing BCE target. References: https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441 rr)rs r3 smooth_bcer$s s?C#I %%r5ceZdZdZiddfd4d Zd5dZd6dZ d7d8dZdZd9d!Z d:d'Z e d()e d;dd3Zd,S)?ConfusionMatrixaA class for calculating and updating a confusion matrix for object detection and classification tasks. Attributes: task (str): The type of task, either 'detect' or 'classify'. matrix (np.ndarray): The confusion matrix, with dimensions depending on the task. nc (int): The number of classes. names (dict[int, str]): The names of the classes, used as labels on the plot. matches (dict | None): Contains the indices of ground truths and predictions categorized into TP, FP and FN. detectFnamesdict[int, str]taskstr save_matchesrc ||_t||_|jdkr tj|j|jfn%tj|jdz|jdzf|_||_|rind|_dS)aNInitialize a ConfusionMatrix instance. Args: names (dict[int, str], optional): Names of classes, used as labels on the plot. task (str, optional): Type of task, either 'detect' or 'classify'. save_matches (bool, optional): Save the indices of GTs, TPs, FPs, FNs for visualization. classifyr8N)rlenncr$zerosmatrixrmatches)selfrrrs r3__init__zConfusionMatrix.__init__?s e**6:i:6M6Mbh1222SUS[]a]dgh]hjnjqtuju\vSwSw  )3rrt r5mtypebatchdict[str, Any]idxintr!Nonec:|jdS|D]|\}}|dvr#|j||xx||gz cc<,|dkrJ|j||xx|dkr|d|dzkgn||gz cc<}dS)aAppend the matches to TP, FP, FN or GT list for the last batch. This method updates the matches dictionary by appending specific batch data to the appropriate match type (True Positive, False Positive, or False Negative). Args: mtype (str): Match type identifier ('TP', 'FP', 'FN' or 'GT'). batch (dict[str, Any]): Batch data containing detection results with keys like 'bboxes', 'cls', 'conf', 'keypoints', 'masks'. idx (int): Index of the specific detection to append from the batch. Notes: For masks, handles both overlap and non-overlap cases. When masks.max() > 1.0, it indicates overlap_mask=True with shape (1, H, W), otherwise uses direct indexing. N>clsconfbboxes keypointsmasksrrr8)ritemsr>)rrrrkrds r3_append_matcheszConfusionMatrix._append_matchesMs <  FKKMM [ [DAq::: U#A&&&!SE(2&&&&g U#A&&&quuww}}1Q437?*;*;STUXSYRZZ&&&  [ [r5predslist[torch.Tensor]targetsc\tj|dddftj|}}t||D] \}}|j||xxdz cc<!dS)zUpdate confusion matrix for classification task. Args: preds (list[torch.Tensor]): Predicted class labels. targets (list[torch.Tensor]): Ground truth class labels. Nrr8)r<rzipcpunumpyr)rrrpts r3process_cls_predsz!ConfusionMatrix.process_cls_predsfs5))!!!Q$/71C1Cw ))++W[[]]-@-@-B-BCC # #DAq KN1    "     # #r5r ? detectionsdict[str, torch.Tensor]rr iou_thresc |d|d}}|jEddD|_t|jdD]}|d|||jdd k}d|rd nd hvrd ndjddk} |jddkr| sfd Dd} t | D]8\}} |j| |jfxxdz cc<|d|9dS| rp|} t | D]8\}} |j|j| fxxdz cc<|d||9dSfdD|} d} d}|rt||nt||}tj ||k}|djdr6tj tj|d||d|dfdddffd}|djddkr||dddfddd}|t%j|dddfdd}||dddfddd}|t%j|dddfdd}nt%jd}|jddk}|t\}}}t | D]\}} ||k}|rt/|dkr| ||} |j| | fxxdz cc<| | kr0|d|||d|||d|||j|j| fxxdz cc<|d|| t | D]K\}} t3||ks3|j| |jfxxdz cc<|d|LdS)aUpdate confusion matrix for object detection task. Args: detections (dict[str, torch.Tensor]): Dictionary containing detected bounding boxes and their associated information. Should contain 'cls', 'conf', and 'bboxes' keys, where 'bboxes' can be Array[N, 4] for regular boxes or Array[N, 5] for OBB with angle. batch (dict[str, Any]): Batch dictionary containing ground truth data with 'bboxes' (Array[M, 4]| Array[M, 5]) and 'cls' (Array[M]) keys, where M is the number of ground truth objects. conf (float, optional): Confidence threshold for detections. iou_thres (float, optional): IoU threshold for matching detections to ground truth. rrNc8i|]}|ttSr)rlist)rrs r3 z1ConfusionMatrix.process_batch..s"SSSQA{400SSSr5>FNFPGTTPrrr8g{Gz?MbP?r cBi|]}||dkSrrrrrrs r3rz1ConfusionMatrix.process_batch..s.^^^aaAz&/AD/H!I^^^r5rrcBi|]}||dkSrrrs r3rz1ConfusionMatrix.process_batch..s.VVVaaAz&'9D'@AVVVr5r9rMT) return_index)rr)rrangeshaperrtolist enumeraterrrrFr<whererstackrrargsortr$uniquer transposeastyperlitemany)rrrrrgt_cls gt_bboxesiis_obbno_preddetection_classesdc gt_classesgcrrrrnm0m1_js ` ` r3 process_batchzConfusionMatrix.process_batchqs$"%L%/  < #SS:RSSSDL6<?++ 5 5$$T5!4444#q(f&?dd%@@@ttdU#)!,1 <?a   >^^^^^S]^^^ $.u$5$9$9$;$;$B$B$D$D!&'899>>EArKDG ,,,1,,,((z1==== F  ,,..J":.. 5 52 DGRK(((A-((($$T5!4444 FVVVVV:VVV ZZ\\((** &u-1133::<<H%28XmIv...giQW>X>X Ki ( ( Q4:a= 'iQ!2!2C!ad OAAAtG4L MqQQUUWW]]__Gtz!}q  !'!!!Q$-"7"7"9"9$$B$"?@!")GAAAqDM"M"M"Ma"PQ!'!!!Q$-"7"7"9"9$$B$"?@!")GAAAqDM"M"M"Ma"PQhv&&G M! q %%''..s33 Bz** 5 5EAraA 5SVVq[[&r!uzz||4 BF###q(###88((z2a5::<<HHHH((z2a5::<<HHH((ua8888 DGRK(((A-((($$T5!4444011 : :EArrQw<< : BK(((A-((($$T:q999 : :r5c|jS)zReturn the confusion matrix.)rrs r3rzConfusionMatrix.matrixs {r5tuple[np.ndarray, np.ndarray]c|j}|jd|z }|jdkr||fn|dd|ddfS)zReturn true positives and false positives. Returns: tp (np.ndarray): True positives. fp (np.ndarray): False positives. r8rNrM)rdiagonalrlr)rtpfps r3tp_fpzConfusionMatrix.tp_fps^[ ! ! # # [__Q  " $9 22BxxCRC"SbS'8JJr5imgr6im_filesave_dirrc 6|jsdSddlm}ddlm}t t }tgdD]\}}|j|} d| vr5tj dgt| dz|j | d<tj t| d|j |z| d <| D]} || xx| | z cc<d |D}|jd kr'|djd r||d|d<|dz dd|||ddddgd|dz t'|jz |jdddS)zPlot grid of GT, TP, FP, FN for each image. Args: img (torch.Tensor): Image to plot onto. im_file (str): Image filename to save visualizations. save_dir (Path): Location to save the visualizations to. Nr8) xyxy2xywh) plot_images)rrrrrrr)rt batch_idxci|]>\}}|t|rtj|dntjd?Sr)rr<remptyrrrds r3rz0ConfusionMatrix.plot_matches..sB```A!#a&&DU[A&&&ek!nn```r5obbrvisualizationsT)parentsexist_okrL)z Ground TruthzFalse PositiveszTrue PositiveszFalse Negativesr)pathsfnamer max_subplots conf_thres)ropsrplottingrrrrr<rurrtoneskeysrrrmkdirrepeatrnamer) rrrrrrlabelsrrmbatchrs r3 plot_matcheszConfusionMatrix.plot_matchess|  F""""""))))))T""!":":":;; ' 'HAu\%(FV##!&sec&:J6K6K.KTWT^!_!_!_v"'*S1A-B-B3:"V"V"VYZ"ZF; [[]] ' 'q VAY&  'a`QWQ]Q]Q_Q_``` 9  &"2"8"; (y)9::F8  $ $++D4+HHH  JJq!Q " "ZZZ--W 0BB*      r5zConfusionMatrix plot failure)msgTN normalizec  ddlm}|j|r1|jddddzndz }t j||dk<|ddd\}}t|j |j } }|j d krWtd |j d z} tdd| } || }|| ddfdd| f}|j | zdz | z} |jd kr| n| dz} d } d| cxkrdkrnn|jd kr|ng|d} | d kr!t jt!| nt j| }tddd| zz }tddd| zz }tddd| zz }tddd| zz }t#j5t#jd||ddd}|jd| dkrd|rdnt j|z}t1|d| D]y\}}t1|d| D]\\}}|||f}t j|r$||||r|dnt7|ddd ||krd!nd"#]z|||d$d%&}dddn #1swxYwYd'd(|zz}|d)|d *|d+|d *|||d,-| ||!||"d.d/d0d/d01|"d2d/d0d/d03| d kr0|#| |d4d5|$| |6d7D]M}|d8kr |j%|&d0|j'j%|&d0N|(dd9d:|;tS||*+dz }|,|d?@|-||r(||dA|j.dBdSdS)CamPlot the confusion matrix using matplotlib and save it to a file. Args: normalize (bool, optional): Whether to normalize the confusion matrix. save_dir (str, optional): Directory where the plot will be saved. on_plot (callable, optional): An optional callback to pass plots path and data when they are rendered. rNr8rM& .>g{Gzt?)r~ )figsizedr9<rautoc backgroundrr~r rignoreBluesnone)cmapvmin interpolationbottomr.2fcenter whiteblack)havafontsizecolorgZd;O?皙?)axfractionpadzConfusion Matrixz NormalizedTrue)rKlabelpad Predicted)rKrPrTF)axisrBtop labelbottomlabeltopy)rUleftright labelleft labelrightZ)rKrotationrI)rK>rVrZr[rBoutliner`gzG?gGz?)rZr[rVrB rz.pngdpiconfusion_matrix)typer)/matplotlib.pyplotpyplotrrlreshaper$nansubplotsrrvaluesrr>slicerarangerwarningscatch_warnings simplefilterimshowxaxisset_label_positionnanmaxrisnantextrcolorbar set_xlabel set_ylabel set_title set_xticks set_yticks tick_paramsset_xticklabelsset_yticklabelsspines set_visiblerNsubplots_adjustrlowerreplacesavefigcloser)rr/ron_plotpltarrayfigrNrrrkeep_idxr ticklabelsxy_ticks tick_fontsizelabel_fontsizetitle_fontsizebtmimcolor_thresholdrrowrvalcbartitles plot_fnames r3plotzConfusionMatrix.plots (''''' Y]  2 2 : :1b A AD H H\]^!veem,,q!W,55R ))++,,dgq 7c>>Atw"}%%AT4++H(OE(AAA+&qqq({3E1qQ&A)z))QQq1u r;;;;B;;;;;"&)z"9"9?U?U ?UJ1;v1E1E29S__---29UW==ArC"H}-- QS2X ..QS2X ..#tebj())  $ & & E E  !( + + +5wSOOB H ' ' 1 1 1Bww"&y*N!!bi>N>N"O'crc 33  FAs"+CH"5"5  3#AqDk8C==%$,5HsLLLc#hh=''%'-0?-B-B''  <<rEt<DDD) E E E E E E E E E E E E E E E*#]Y%>> f~ CCC kNR HHH U^ <<< h h C%TTYZZZ Cd%4TYZZZ      zMBS[  \ \ \   zM  B B B> 1 1AI~~ ! ((/// GN1  ) )% 0 0 0 0 $DEEE(^^)>)>sC)H)H&N&N&NN  JC ((( #  ^ GJ);t{GYGYG[G[ \ \ ] ] ] ] ] ^ ^sDK==LLc t|jjdD]G}tjdt t|j|HdS)z*Print the confusion matrix to the console.rraN)rrrrinfojoinmaprrrs r3printzConfusionMatrix.print@s\t{(+,, < >> results = model.val(data="coco8.yaml", plots=True) >>> cm_dict = results.confusion_matrix.summary(normalize=True, decimals=5) >>> print(cm_dict) rNrr8z [^a-zA-Z0-9_]rr8rMr1c g|]?tdififdttD@S)rSc2i|]}||fSrr)rrr clean_namesrs r3rz6ConfusionMatrix.summary...fs(2p2p2pST;q>5A;2p2p2pr5)dictrr)rrrrs @r3 z+ConfusionMatrix.summary..esq    +{1~. q q2p2p2p2p2p2pX]^abm^n^nXoXo2p2p2p q q   r5)rerrrrlsetsubraddappendrrlriroundrr) rr/rrrseenr) clean_nameoriginal_cleancounterrrs @@r3summaryzConfusionMatrix.summaryEs$  -1Y*-D-DTZ&&(()))JtDQUQ[QbQbQdQdLeLeJtgsJt T + +D 0#t<argmaxryrzrrrr{rrrr) rrrrrrrrrrNrrYs r3 plot_mc_curvers *$#####ll1adlCCGC3u::bMM = =DAq GGBQqmG < < < < = BDAV444rwwqzz3AGGBQf4h155774h4h4hTVWXW_W_WaWaTb4h4h4hGiiiMM&MM&KK1KK1IIYLI999LLF---...KKcK"""IIcNNNcfllnn#<#<#<299;;UWU^U^U`U`aabbbbbccr5recall precision$tuple[float, np.ndarray, np.ndarray]ctjdg|dgf}tjdg|dgf}tjtjtj|}d}|dkrntjddd}t jtjdr tj n tj }|tj ||||}n`tj |dd|dd kd}tj ||dz||z ||dzz}|||fS) aCompute the average precision (AP) given the recall and precision curves. Args: recall (list[float]): The recall curve. precision (list[float]): The precision curve. Returns: ap (float): Average precision. mpre (np.ndarray): Precision envelope curve. mrec (np.ndarray): Modified recall curve with sentinel values added at the beginning and end. r=rinterprr8ez>=2.0NrM)r$rflipr& accumulatelinspacer check_version __version__ trapezoidtrapzrrrl) rrmrecmpremethodrfuncrrs r3 compute_aprs, >C5&3%0 1 1D >C5)cU3 4 4D 72:((77 8 8DF  K1c " "%3BNGLLZr||RTRZ T")AtT**A . . HT!""Xcrc* + +A . VT!a%[47*d1q5k9 : : tT>r5gؗҜCompute the average precision per class for object detection evaluation. Args: tp (np.ndarray): Binary array indicating whether the detection is correct (True) or not (False). conf (np.ndarray): Array of confidence scores of the detections. pred_cls (np.ndarray): Array of predicted classes of the detections. target_cls (np.ndarray): Array of true classes of the targets. plot (bool, optional): Whether to plot PR curves or not. on_plot (callable, optional): A callback to pass plots path and data when they are rendered. save_dir (Path, optional): Directory to save the PR curves. names (dict[int, str], optional): Dictionary of class names to plot PR curves. eps (float, optional): A small value to avoid division by zero. prefix (str, optional): A prefix string for saving the plot files. Returns: tp (np.ndarray): True positive counts at threshold given by max F1 metric for each class. fp (np.ndarray): False positive counts at threshold given by max F1 metric for each class. p (np.ndarray): Precision values at threshold given by max F1 metric for each class. r (np.ndarray): Recall values at threshold given by max F1 metric for each class. f1 (np.ndarray): F1-score values at threshold given by max F1 metric for each class. ap (np.ndarray): Average precision for each class at different IoU thresholds. unique_classes (np.ndarray): An array of unique classes that have data. p_curve (np.ndarray): Precision curves for each class. r_curve (np.ndarray): Recall curves for each class. f1_curve (np.ndarray): F1-score curves for each class. x (np.ndarray): X-axis values for the curves. prec_values (np.ndarray): Precision values at mAP@0.5 for each class. T) return_countsrr8N)rZ)r8rr9c0i|]\}}|v ||Srr)rrrrs r3rz ap_per_class..@s( M M MTQ!u**Qa***r5z PR_curve.pngrz F1_curve.pngF1)rrz P_curve.pngrz R_curve.pngrr)r$rrrrrrrlcumsumrrrrrrrrrrrrr)#r rrrrrrrrrrunique_classesntrr prec_valuesrp_curver_curvecirn_ln_pfpctpcrrrrrf1_curverrf1rs# ` r3 ap_per_classrs R D5AAQ!hB:TBBBNB  a B[At,,b{A8R!$566"d8L8LbhXZ\`WaNbNbB>**==A Mfeegg !88saxx 2a5y  ##ell1ooc "iT!WHfQQQTlCCC 39% iT!WHi1oAFFF rx{## = =A$.vaaad|Yqqq!t_$M$M !Br1uItTAvv""29Qd#;#;<<< = ,7O"(;'''BHY%> M M ME nab(5L5L5L*Le]deeeea8.E.E.E#EuUYcjkkkka(-C-C-C"CUS^hoppppa(-C-C-C"CUS[elmmmmx}}Q%%,,..Aqqq!t}gaaadmXaaad^"qA b&  B C.2  $ $ & &B r1aR!6!6s!;!;WgxYZ\g ggr5c*eZdZdZddZed dZed dZed!d Zed!d Z ed!d Z ed!d Z ed!d Z d"dZ d#dZed$dZd!dZd%dZed&dZed'dZdS)(raClass for computing evaluation metrics for Ultralytics YOLO models. Attributes: p (list): Precision for each class. Shape: (nc,). r (list): Recall for each class. Shape: (nc,). f1 (list): F1 score for each class. Shape: (nc,). all_ap (list): AP scores for all classes and all IoU thresholds. Shape: (nc, 10). ap_class_index (list): Index of class for each AP score. Shape: (nc,). nc (int): Number of classes. Methods: ap50: AP at IoU threshold of 0.5 for all classes. ap: AP at IoU thresholds from 0.5 to 0.95 for all classes. mp: Mean precision of all classes. mr: Mean recall of all classes. map50: Mean AP at IoU threshold of 0.5 for all classes. map75: Mean AP at IoU threshold of 0.75 for all classes. map: Mean AP at IoU thresholds from 0.5 to 0.95 for all classes. mean_results: Mean of results, returns mp, mr, map50, map. class_result: Class-aware result, returns p[i], r[i], ap50[i], ap[i]. maps: mAP of each class. fitness: Model fitness as a weighted combination of metrics. update: Update metric attributes with new evaluation results. curves: Provides a list of curves for accessing specific metrics like precision, recall, F1, etc. curves_results: Provide a list of results for accessing specific metrics like precision, recall, F1, etc. r!rcZg|_g|_g|_g|_g|_d|_dS)zQInitialize a Metric instance for computing evaluation metrics for the YOLO model.rN)rrrall_apap_class_indexrr s r3rzMetric.__init__js1  r5np.ndarray | listcPt|jr|jdddfngS)zReturn the Average Precision (AP) at an IoU threshold of 0.5 for all classes. Returns: (np.ndarray | list): Array of shape (nc,) with AP50 values per class, or an empty list if not available. Nr)rrr s r3ap50z Metric.ap50ss-%( $4$4t{"""B>r5r c`t|jr|jndS)z|Return the Mean Precision of all classes. Returns: (float): The mean precision of all classes. r=)rrrr s r3mpz Metric.mp%!$DF 4tv{{}}}4r5c`t|jr|jndS)zvReturn the Mean Recall of all classes. Returns: (float): The mean recall of all classes. r=)rrrr s r3mrz Metric.mrrr5ctt|jr#|jdddfndS)zReturn the mean Average Precision (mAP) at an IoU threshold of 0.5. Returns: (float): The mAP at an IoU threshold of 0.5. Nrr=rr s r3map50z Metric.map509,/t{+;+;Dt{111a4 %%'''Dr5ctt|jr#|jdddfndS)zReturn the mean Average Precision (mAP) at an IoU threshold of 0.75. Returns: (float): The mAP at an IoU threshold of 0.75. Nrr=rr s r3map75z Metric.map75r$r5c`t|jr|jndS)zReturn the mean Average Precision (mAP) over IoU thresholds of 0.5 - 0.95 in steps of 0.05. Returns: (float): The mAP over IoU thresholds of 0.5 - 0.95 in steps of 0.05. r=rr s r3rz Metric.maps,&)%5%5>t{!!!3>r5rrc6|j|j|j|jgS)z+Return mean of results, mp, mr, map50, map.)rr!r#rr s r3 mean_resultszMetric.mean_resultss$*dh77r5rr!tuple[float, float, float, float]cf|j||j||j||j|fS)z6Return class-aware result, p[i], r[i], ap50[i], ap[i].)rrrrrs r3 class_resultzMetric.class_results)vay$&)TYq\471:==r5rctj|j|jz}t |jD]\}}|j|||<|S)zReturn mAP of each class.)r$rrrrrr)rmapsrrs r3r.z Metric.mapssPx  48+d122 ! !DAqgajDGG r5cgd}ttjtj||zS)z:Return model fitness as a weighted combination of metrics.)r=r=r=r)r r$ nan_to_numrr)rl)rws r3fitnesszMetric.fitnesssL bmBHT->->-@-@$A$ABBQFKKMMNNNr5resultsrc |\ |_|_|_|_|_|_|_|_|_|_ dS)a Update the evaluation metrics with a new set of results. Args: results (tuple): A tuple containing evaluation metrics: - p (list): Precision for each class. - r (list): Recall for each class. - f1 (list): F1 score for each class. - all_ap (list): AP scores for all classes and all IoU thresholds. - ap_class_index (list): Index of class for each AP score. - p_curve (list): Precision curve for each class. - r_curve (list): Recall curve for each class. - f1_curve (list): F1 curve for each class. - px (list): X values for the curves. - prec_values (list): Precision values for each class. N) rrrrrr r rrr)rr3s r3updatez Metric.updatesF6  F F G K   L L M G    r5rcgS>Return a list of curves for accessing specific metrics curves.rr s r3curvesz Metric.curves  r5 list[list]c~|j|jddg|j|jddg|j|jddg|j|jddggS)FReturn a list of curves results for accessing specific metrics curves.rrrr)rrrr r r s r3curves_resultszMetric.curves_resultssOWd&+ > Wdm\4 8 WdlL+ > WdlL( ;   r5Nr!r)r!rr!r r!rrrrr!r*r!r)r3rr!rr!r;)rrrrrpropertyrrrr!r#r&rr)r,r.r2r5r9r>rr5r3rrNs6===X=???X?555X5555X5EEEXEEEEXE???X?8888>>>>XOOOO :X   X   r5ceZdZdZifd-dZd.d Zed d d fd/dZdZe d0dZ d1dZ d2dZ e d3dZ e d4dZe d5d!Ze d6d#Ze d0d$Ze d7d&Zd8d9d,Zd S): DetMetricsaUtility class for computing detection metrics such as precision, recall, and mean average precision (mAP). Attributes: names (dict[int, str]): A dictionary of class names. box (Metric): An instance of the Metric class for storing detection results. speed (dict[str, float]): A dictionary for storing execution times of different parts of the detection process. stats (dict[str, list]): A dictionary containing lists for true positives, confidence scores, predicted classes, target classes, and target images. nt_per_class: Number of targets per class. nt_per_image: Number of targets per image. Methods: update_stats: Update statistics by appending new values to existing stat collections. process: Process predicted results for object detection and update metrics. clear_stats: Clear the stored statistics. keys: Return a list of keys for accessing specific metrics. mean_results: Calculate mean of detected objects & return precision, recall, mAP50, and mAP50-95. class_result: Return the result of evaluating the performance of an object detection model on a specific class. maps: Return mean Average Precision (mAP) scores per class. fitness: Return the fitness of box object. ap_class_index: Return the average precision index per class. results_dict: Return dictionary of computed performance metrics and statistics. curves: Return a list of curves for accessing specific metrics curves. curves_results: Return a list of computed performance metrics and statistics. summary: Generate a summarized representation of per-class detection metrics as a list of dictionaries. rrr!rc||_t|_ddddd|_t ggggg|_d|_d|_dS)zInitialize a DetMetrics instance with class names. Args: names (dict[int, str], optional): Dictionary of class names. r= preprocess inferenceloss postprocess)r rrr target_imgN)rrboxspeedrstats nt_per_class nt_per_imagerrs r3rzDetMetrics.__init__s[  88$'c3WZ[[ Rb2"QSTTT   r5statrc|jD](}|j|||)dS)zUpdate statistics by appending new values to existing stat collections. Args: stat (dict[str, Any]): Dictionary containing new statistical values to append. Keys should match existing keys in self.stats. N)rRr&r)rrVrs r3 update_statszDetMetrics.update_statssJ"" * *A JqM a ) ) ) ) * *r5.FNrrrrdict[str, np.ndarray]c jd|jD}|s|St|d|d|d|d|||j|d dd }t |j|j_|j|tj |d tt |j |_ tj |d  tt |j |_ |S) aProcess predicted results for object detection and update metrics. Args: save_dir (Path): Directory to save plots. Defaults to Path("."). plot (bool): Whether to plot precision-recall curves. Defaults to False. on_plot (callable, optional): Function to call after plots are generated. Defaults to None. Returns: (dict[str, np.ndarray]): Dictionary containing concatenated statistics arrays. c@i|]\}}|tj|dSr)r$rrs r3rz&DetMetrics.process..2s*HHHTQBN1a((HHHr5r rrrBox)rrrrrr9N) minlengthrO)rRrrrrrPrr5r$bincountrrrSrT)rrrrrRr3s r3processzDetMetrics.process'sIHTZ5E5E5G5GHHH L $K &M *  , *     "" $*oo     Kl(;(B(B3(G(GSVW[WaSbSbcccKl(;(B(B3(G(GSVW[WaSbSbccc r5cf|jD]}|dS)zClear the stored statistics.N)rRrlclear)rrds r3 clear_statszDetMetrics.clear_statsFs8""$$  A GGIIII  r5 list[str]c gdS)z5Return a list of keys for accessing specific metrics.)zmetrics/precision(B)zmetrics/recall(B)zmetrics/mAP50(B)zmetrics/mAP50-95(B)rr s r3r&zDetMetrics.keysKshgggr5rrc4|jS)zSCalculate mean of detected objects & return precision, recall, mAP50, and mAP50-95.)rPr)r s r3r)zDetMetrics.mean_resultsPsx$$&&&r5rrr*c6|j|S)zaReturn the result of evaluating the performance of an object detection model on a specific class.)rPr,rs r3r,zDetMetrics.class_resultTsx$$Q'''r5rc|jjS)z5Return mean Average Precision (mAP) scores per class.)rPr.r s r3r.zDetMetrics.mapsXsx}r5r c4|jS)z!Return the fitness of box object.)rPr2r s r3r2zDetMetrics.fitness]sx!!!r5rc|jjS)z-Return the average precision index per class.)rPrr s r3rzDetMetrics.ap_class_indexbx&&r5dict[str, float]cg|jd}dg||jD}tt ||S)zAReturn dictionary of computed performance metrics and statistics.r2c3XK|]%}t|drt|n|V&dS)rN)hasattrr rs r3rz*DetMetrics.results_dict..ks;jjawq&1185888qjjjjjjr5)r&r)r2rr)rr&rls r3 results_dictzDetMetrics.results_dictgs['&I&jjDhdFWFWFYFYDh[_[gDhjjjCf%%&&&r5c gdS)r8)Precision-Recall(B)F1-Confidence(B)Precision-Confidence(B)Recall-Confidence(B)rr s r3r9zDetMetrics.curvesnsnmmmr5r;c|jjSz=Return a list of computed performance metrics and statistics.)rPr>r s r3r>zDetMetrics.curves_resultssrkr5Trr/rlist[dict[str, Any]]cjjjjjjdfdt t dDS)aGenerate a summarized representation of per-class detection metrics as a list of dictionaries. Includes shared scalar metrics (mAP, mAP50, mAP75) alongside precision, recall, and F1-score for each class. Args: normalize (bool): For Detect metrics, everything is normalized by default [0-1]. decimals (int): Number of decimal places to round the metrics values to. Returns: (list[dict[str, Any]]): A list of dictionaries, each representing one class with corresponding metric values. Examples: >>> results = model.val(data="coco8.yaml") >>> detection_summary = results.summary() >>> print(detection_summary) )Box-PzBox-RzBox-F1c g|]jjjjjjdfdDt dt ddS))ClassImages InstancescDi|]\}}|t|Srrrrrdrrs r3rz1DetMetrics.summary...s-JJJ11eAaD(++JJJr5r9r)mAP50zmAP50-95)rrrTrSrrr,)rrr per_classrs @r3rz&DetMetrics.summary..s    D$7$:;+D,?,BC!.t/B1/EF  KJJJJ 8I8IJJJ   t0033A6AA!$"3"3A"6"6q"98DD        r5rz)rPrrrrr)rr/rrs` `@r3rzDetMetrics.summaryxst$XZXZhk        3y12233    r5rrr!r)rVrr!rrrrrr!rZr!rdrArBrCr@rDr!rlrETrr/rrrr!rx)rrrrrrXrr`rcrFr&r)r,r.r2rrpr9r>rrr5r3rHrHs602 ! ! ! ! !****(,tCyyud> hhhXh''''((((X"""X"'''X''''X' nnnXn'''X'        r5rHceZdZdZifd$dZeddd fd%dZed&dZd'dZ d(dZ ed)dZ ed*dZ ed&dZ ed+dZd,d-d#Zd S).SegmentMetricsaCalculate and aggregate detection and segmentation metrics over a given set of classes. Attributes: names (dict[int, str]): Dictionary of class names. box (Metric): An instance of the Metric class for storing detection results. seg (Metric): An instance of the Metric class to calculate mask segmentation metrics. speed (dict[str, float]): A dictionary for storing execution times of different parts of the detection process. stats (dict[str, list]): A dictionary containing lists for true positives, confidence scores, predicted classes, target classes, and target images. nt_per_class: Number of targets per class. nt_per_image: Number of targets per image. Methods: process: Process the detection and segmentation metrics over the given set of predictions. keys: Return a list of keys for accessing metrics. mean_results: Return the mean metrics for bounding box and segmentation results. class_result: Return classification results for a specified class index. maps: Return mAP scores for object detection and segmentation models. fitness: Return the fitness score for both segmentation and bounding box models. curves: Return a list of curves for accessing specific metrics curves. curves_results: Provide a list of computed performance metrics and statistics. summary: Generate a summarized representation of per-class segmentation metrics as a list of dictionaries. rrr!rcvt||t|_g|jd<dS)zInitialize a SegmentMetrics instance with class names. Args: names (dict[int, str], optional): Dictionary of class names. tp_mN)rHrrsegrRrUs r3rzSegmentMetrics.__init__s7 D%(((88 6r5rYFNrrrrrZc ,t||||}t|d|d|d|d||||jd dd }t |j|j_|j||S) aProcess the detection and segmentation metrics over the given set of predictions. Args: save_dir (Path): Directory to save plots. Defaults to Path("."). plot (bool): Whether to plot precision-recall curves. Defaults to False. on_plot (callable, optional): Function to call after plots are generated. Defaults to None. Returns: (dict[str, np.ndarray]): Dictionary containing concatenated statistics arrays. rrrrrMaskrrrrrr9N)rHr`rrrrrr5)rrrrrR results_masks r3r`zSegmentMetrics.processs""44"II# &M &M *  , *     ""  $*oo   %%% r5rdcTgtj|ddddS)z,Return a list of keys for accessing metrics.zmetrics/precision(M)zmetrics/recall(M)zmetrics/mAP50(M)zmetrics/mAP50-95(M)rHr&fgetr s r3r&zSegmentMetrics.keysJ _ ! !$ ' ' "      "   r5rrcjt||jzS)zBReturn the mean metrics for bounding box and segmentation results.)rHr)rr s r3r)zSegmentMetrics.mean_resultss)&&t,,tx/D/D/F/FFFr5rrcnt|||j|zS)z:Return classification results for a specified class index.)rHr,rrs r3r,zSegmentMetrics.class_results-&&tQ//$(2G2G2J2JJJr5rcZtj||jjzS)z?Return mAP scores for object detection and segmentation models.)rHr.rrr s r3r.zSegmentMetrics.mapss###D))DHM99r5r ct|jtj|zS)zGReturn the fitness score for both segmentation and bounding box models.)rr2rHrr s r3r2zSegmentMetrics.fitnesss.x!!J$6$;$;D$A$AAAr5cTgtj|ddddS)r8zPrecision-Recall(M)zF1-Confidence(M)zPrecision-Confidence(M)zRecall-Confidence(M)rHr9rr s r3r9zSegmentMetrics.curvessK   # #D ) ) !    &  #   r5r;cZtj||jjzSrw)rHr>rrr s r3r>zSegmentMetrics.curves_resultss%(--d33dh6MMMr5Trr/rrxc |jj|jj|jjd}t||}t |D];\}|ifd|D<|S)aGenerate a summarized representation of per-class segmentation metrics as a list of dictionaries. Includes both box and mask scalar metrics (mAP, mAP50, mAP75) alongside precision, recall, and F1-score for each class. Args: normalize (bool): For Segment metrics, everything is normalized by default [0-1]. decimals (int): Number of decimal places to round the metrics values to. Returns: (list[dict[str, Any]]): A list of dictionaries, each representing one class with corresponding metric values. Examples: >>> results = model.val(data="coco8-seg.yaml") >>> seg_summary = results.summary(decimals=4) >>> print(seg_summary) )zMask-PzMask-RzMask-F1cDi|]\}}|t|Srrrs r3rz*SegmentMetrics.summary..-PPPdaE!A$11PPPr5) rrrrrHrrr5rrr/rrrrrs ` @r3rzSegmentMetrics.summarys&hjhjx{  $$T9h??g&& S SDAq HHQPPPPPioo>O>OPPPQ R R R Rr5rrrrArrr!rrrCr@rErr)rrrrrrr`rFr&r)r,r.r2r9r>rrr5r3rrsW002     (,tCyyud6   X GGGGKKKK:::X:BBBXB   X NNNXNr5rceZdZdZifd$fd Zeddd fd%dZed&dZd'dZ d(dZ ed)dZ ed*dZ ed&dZ ed+dZd,d-d#ZxZS). PoseMetricsaCalculate and aggregate detection and pose metrics over a given set of classes. Attributes: names (dict[int, str]): Dictionary of class names. pose (Metric): An instance of the Metric class to calculate pose metrics. box (Metric): An instance of the Metric class for storing detection results. speed (dict[str, float]): A dictionary for storing execution times of different parts of the detection process. stats (dict[str, list]): A dictionary containing lists for true positives, confidence scores, predicted classes, target classes, and target images. nt_per_class: Number of targets per class. nt_per_image: Number of targets per image. Methods: process: Process the detection and pose metrics over the given set of predictions. keys: Return a list of keys for accessing metrics. mean_results: Return the mean results of box and pose. class_result: Return the class-wise detection results for a specific class i. maps: Return the mean average precision (mAP) per class for both box and pose detections. fitness: Return combined fitness score for pose and box detection. curves: Return a list of curves for accessing specific metrics curves. curves_results: Provide a list of computed performance metrics and statistics. summary: Generate a summarized representation of per-class pose metrics as a list of dictionaries. rrr!rct|t|_g|jd<dS)zInitialize the PoseMetrics class with class names. Args: names (dict[int, str], optional): Dictionary of class names. tp_pN)superrrposerR)rr __class__s r3rzPoseMetrics.__init__<s: HH  6r5rYFNrrrrrZc ,t||||}t|d|d|d|d||||jd dd }t |j|j_|j||S) aProcess the detection and pose metrics over the given set of predictions. Args: save_dir (Path): Directory to save plots. Defaults to Path("."). plot (bool): Whether to plot precision-recall curves. Defaults to False. on_plot (callable, optional): Function to call after plots are generated. Returns: (dict[str, np.ndarray]): Dictionary containing concatenated statistics arrays. rrrrrPoserr9N)rHr`rrrrrr5)rrrrrR results_poses r3r`zPoseMetrics.processFs""44"II# &M &M *  , *     ""  4:  &&& r5rdcTgtj|ddddS)z(Return a list of evaluation metric keys.zmetrics/precision(P)zmetrics/recall(P)zmetrics/mAP50(P)zmetrics/mAP50-95(P)rr s r3r&zPoseMetrics.keysarr5rrcjt||jzS)z(Return the mean results of box and pose.)rHr)rr s r3r)zPoseMetrics.mean_resultsls)&&t,,ty/E/E/G/GGGr5rrcnt|||j|zS)z?Return the class-wise detection results for a specific class i.)rHr,rrs r3r,zPoseMetrics.class_resultps-&&tQ//$)2H2H2K2KKKr5rcZtj||jjzS)zSReturn the mean average precision (mAP) per class for both box and pose detections.)rHr.rrr s r3r.zPoseMetrics.mapsts###D))DIN::r5r ct|jtj|zS)z9Return combined fitness score for pose and box detection.)rr2rHrr s r3r2zPoseMetrics.fitnessys.y  ""Z%7%<%rrr s r3r>zPoseMetrics.curves_resultss%(--d33di6NNNr5Trr/rrxc |jj|jj|jjd}t||}t |D];\}|ifd|D<|S)aGenerate a summarized representation of per-class pose metrics as a list of dictionaries. Includes both box and pose scalar metrics (mAP, mAP50, mAP75) alongside precision, recall, and F1-score for each class. Args: normalize (bool): For Pose metrics, everything is normalized by default [0-1]. decimals (int): Number of decimal places to round the metrics values to. Returns: (list[dict[str, Any]]): A list of dictionaries, each representing one class with corresponding metric values. Examples: >>> results = model.val(data="coco8-pose.yaml") >>> pose_summary = results.summary(decimals=4) >>> print(pose_summary) )zPose-PzPose-RzPose-F1cDi|]\}}|t|Srrrs r3rz'PoseMetrics.summary..rr5) rrrrrHrrr5rrs ` @r3rzPoseMetrics.summarys$ikiky|  $$T9h??g&& S SDAq HHQPPPPPioo>O>OPPPQ R R R Rr5rrrrArrCr@rErr)rrrrrrr`rFr&r)r,r.r2r9r>r __classcell__)rs@r3rr#sl002       (,tCyyud6   X HHHHLLLL;;;X;CCCXC    X  OOOXOr5rceZdZdZddZddZedd Zedd ZeddZ ed dZ ed dZ d!d"dZ dS)#ClassifyMetricsayClass for computing classification metrics including top-1 and top-5 accuracy. Attributes: top1 (float): The top-1 accuracy. top5 (float): The top-5 accuracy. speed (dict[str, float]): A dictionary containing the time taken for each step in the pipeline. Methods: process: Process target classes and predicted classes to compute metrics. fitness: Return mean of top-1 and top-5 accuracies as fitness score. results_dict: Return a dictionary with model's performance metrics and fitness score. keys: Return a list of keys for the results_dict property. curves: Return a list of curves for accessing specific metrics curves. curves_results: Provide a list of computed performance metrics and statistics. summary: Generate a single-row summary of classification metrics (Top-1 and Top-5 accuracy). r!rc:d|_d|_ddddd|_dS)z&Initialize a ClassifyMetrics instance.rr=rJN)top1top5rQr s r3rzClassifyMetrics.__init__s(  $'c3WZ[[ r5rr6predcvtj|tj|}}|dddf|k}tj|dddf|djfd}|d\|_|_ dS)zProcess target classes and predicted classes to compute metrics. Args: targets (torch.Tensor): Target classes. pred (torch.Tensor): Predicted classes. Nrr8r) r<rr rr>rlrrrr)rrrcorrectaccs r3r`zClassifyMetrics.processs $7););g111d7#t+2244k7111a4='++a..*?@aHHH"xx{{1133 4999r5r c&|j|jzdz S)z;Return mean of top-1 and top-5 accuracies as fitness score.r9)rrr s r3r2zClassifyMetrics.fitnesss DI%**r5rlcrttg|jd|j|j|jgS)zGReturn a dictionary with model's performance metrics and fitness score.r2)rrr&rrr2r s r3rpzClassifyMetrics.results_dicts6C/$)/Y/$)TY 1UVVWWWr5rdc ddgS)z4Return a list of keys for the results_dict property.zmetrics/accuracy_top1zmetrics/accuracy_top5rr s r3r&zClassifyMetrics.keyss()@AAr5rcgSr7rr s r3r9zClassifyMetrics.curvesr:r5cgS)r=rr s r3r>zClassifyMetrics.curves_resultsr:r5Trr/rrrrcZt|j|t|j|dgS)aWGenerate a single-row summary of classification metrics (Top-1 and Top-5 accuracy). Args: normalize (bool): For Classify metrics, everything is normalized by default [0-1]. decimals (int): Number of decimal places to round the metrics values to. Returns: (list[dict[str, float]]): A list with one dictionary containing Top-1 and Top-5 classification accuracy. Examples: >>> results = model.val(data="imagenet10") >>> classify_summary = results.summary(decimals=4) >>> print(classify_summary) )top1_acctop5_acc)rrr)rr/rs r3rzClassifyMetrics.summarys.#49h77U49V^E_E_``aar5Nr?)rr6rr6r@rrrDrr) rrrrrr`rFr2rpr&r9r>rrr5r3rrs"\\\\ 4 4 4 4+++X+XXXXXBBBXBXXbbbbbbbr5rceZdZdZifddZdS) OBBMetricsaMetrics for evaluating oriented bounding box (OBB) detection. Attributes: names (dict[int, str]): Dictionary of class names. box (Metric): An instance of the Metric class for storing detection results. speed (dict[str, float]): A dictionary for storing execution times of different parts of the detection process. stats (dict[str, list]): A dictionary containing lists for true positives, confidence scores, predicted classes, target classes, and target images. nt_per_class: Number of targets per class. nt_per_image: Number of targets per image. References: https://arxiv.org/pdf/2106.06072.pdf rrr!rc<t||dS)zInitialize an OBBMetrics instance with class names. Args: names (dict[int, str], optional): Dictionary of class names. N)rHrrUs r3rzOBBMetrics.__init__ s D%(((((r5Nr)rrrrrrr5r3rrs<  02)))))))r5r)Fr) rrrrrrrr r!r)r)rr6rr6rr r!r6)TFFFr)rr6rr6rGrrHrrIrrJrrr r!r6)rhr6rir6rr r!r6) ror6rpr6r1r6rqrrrr r!r6)r{r6r!r|) rr6rr6rJrrr r!r6)rrrrrr r!r6)r)rr r!r)rM)rYrrr r!r) rrrrrrrrrr) rrrrrrrrrrrr)rrrrrrr!r)r rrrrrrrrrrrrrrr rrr!r).r __future__rrOro collectionsrpathlibrtypingrrr$r<ultralytics.utilsrrr r r r rfloat32 OKS_SIGMA RLE_WEIGHTr4rFrgrnrzrrrrrrrrrrrrHrrrrrr5r3rss"""""" ###### cccccccccccccccc BHnnnj   RXkkk l l %%%%%>IIIII4BBBBBJ(((((&bfXXXXX.IIII(*****Z!!!!!H & & & & & t t t t t ot t t n ;;;;; T.))  )i)i)i)i)iXT.)) *c*c*c*c*cZJ TVV]h]h]h]h]h@c c c c c [c c c Ld d d d d od d d NEEEEEZEEEPHHHHH*HHHVLbLbLbLbLbk?LbLbLb^))))))))))r5