FPh)ddlmZddlZddlZddlmZddlmZm Z ddl m Z ddl m Z mZmZmZddlmZmZGdd eZy) )PathN)DetectionValidator)LOGGERops)check_requirements) OKS_SIGMA PoseMetricsbox_ioukpt_iou)output_to_target plot_imagescjeZdZdZd fd ZfdZdZdZfdZdZ ddZ d Z d Z d Z d ZxZS) PoseValidatoraP A class extending the DetectionValidator class for validation based on a pose model. Example: ```python from ultralytics.models.yolo.pose import PoseValidator args = dict(model='yolov8n-pose.pt', data='coco8-pose.yaml') validator = PoseValidator(args=args) validator() ``` czt||||||d|_d|_d|j_t |j|j|_ t|jjtr>|jjjdk(rtjdyyy)zSInitialize a 'PoseValidator' object with custom parameters and assigned attributes.Npose)save_diron_plotmpsuWARNING ⚠️ Apple MPS known Pose bug. Recommend 'device=cpu' for Pose models. See https://github.com/ultralytics/ultralytics/issues/4031.)super__init__sigma kpt_shapeargstaskr rrmetrics isinstancedevicestrlowerrwarning)self dataloaderrpbarr _callbacks __class__s kC:\Users\daisl\Desktop\realtime-object-detection\venv\Lib\site-packages\ultralytics/models/yolo/pose/val.pyrzPoseValidator.__init__s XtT:F  "DMM4<<P dii&& ,1A1A1G1G1IU1R NNY Z2S ,ct||}|dj|jj |d<|S)zcPreprocesses the batch by converting the 'keypoints' data into a float and moving it to the device. keypoints)r preprocesstorfloat)r!batchr%s r&r*zPoseValidator.preprocess(s>"5)";/224;;?EEGk r'c ddzS)z;Returns description of evaluation metrics in string format.z,%22s%11s%11s%11s%11s%11s%11s%11s%11s%11s%11s) ClassImages InstanceszBox(PRmAP50 mAP50-95)zPose(Pr2r3r4)r!s r&get_desczPoseValidator.get_desc.s$)DD Dr'c tj||jj|jj|j d|jj |jj|jS)zPApply non-maximum suppression and return detections with high confidence scores.T)labels multi_labelagnosticmax_detnc) rnon_max_suppressionrconfioulb single_clsr;r<)r!predss r& postprocesszPoseValidator.postprocess3sX&&u'+yy~~'+yy}}.2gg3704 0D0D/3yy/@/@*.''3 3r'ct|||jd|_|jddgk(}|jd}|r t|_yt j ||z |_y)z0Initiate pose estimation metrics for YOLO model.rrN)r init_metricsdatarrnponesr)r!modelis_posenkptr%s r&rGzPoseValidator.init_metrics>s\ U#;/..RG+~~a ")Y rwwt}t/C r'c 4t|D] \}}|d|k(}|d|}|d|}|d|}|jd|jd} } |jd} |d|} tj| |jtj |j } tj| |jtj |j }|xjdz c_| dk(r| r|jj|| gtjd |j |jd |jjr,|jjd |jd  |jjr d|d d df<|j!}t#j$|d|jdd |d d d df| |d||d d dd fj'| | d }t#j(|d|jdd || |d|| r`|djdd \}}t#j*|tj,||||f|j z}t#j$|d|jdd || |d||j!}|dxx|zcc<|dxx|zcc<t#j(|d|jdd || |d|}tj.||fd}|j1|d d d df|}|j1|d d d df|||} |jjr|jj|||jj|| |d d df|d d df|jd f|jj2s|j5||d| y )zMetrics. batch_idxclsbboxesr)r ori_shape)dtyper)r)rN) detectionsr8img ratio_pad)r[rU).r).rRim_file) enumerateshapetorchzerosniouboolrseenstatsappendsqueezerplotsconfusion_matrix process_batchrAcloner scale_boxesview scale_coords xywh2xyxytensorcat_process_batch save_json pred_to_json)r!rBr-sipredidxrPbboxkptsnlnprnkr_ correct_kptscorrect_bboxespredn pred_kptsheightwidthtboxtkptslabelsns r&update_metricszPoseValidator.update_metricsFs!%(HB $*C,s#C?3'D%c*DiilDJJqMBAB+&r*E ;;sDIIUZZPTP[P[\L"[[diiuzzRVR]R]^N IINIaxJJ%%~|'Fekkt{{G4'F58[[_'FGyy--;;tTWT_T_`bTc;dyy##QT JJLE OOE%L,22126a!e e&+K&8&< >ae ))#r26I   U5\"-33AB7EUZ[fUghjUk l %e 2 212 6 }}T*U\\FE624;;.HHe R 0 6 6qr :D%*/ *g)r'c.|U|Stj|ddddfddddfjddz}t|||j|}nt |ddddf|ddddf}|j |dddf|dddf|S) a Return correct prediction matrix. Args: detections (torch.Tensor): Tensor of shape [N, 6] representing detections. Each detection is of the format: x1, y1, x2, y2, conf, class. labels (torch.Tensor): Tensor of shape [M, 5] representing labels. Each label is of the format: class, x1, y1, x2, y2. pred_kpts (torch.Tensor, optional): Tensor of shape [N, 51] representing predicted keypoints. 51 corresponds to 17 keypoints each with 3 values. gt_kpts (torch.Tensor, optional): Tensor of shape [N, 51] representing ground truth keypoints. Returns: torch.Tensor: Correct prediction matrix of shape [N, 10] for 10 IoU levels. NrRrUg(\?)rarearZrXr)r xyxy2xywhprodr rr match_predictions)r!rWr8rgt_kptsrr?s r&rrzPoseValidator._process_batchs  W%8==12/126;;A>ED'9DJJTJC&AB-ArrE):;C%%jA&6q!t cJJr'c t|d|d|djd|d|d|d|jd|d z |j|j y ) zSPlots and saves validation set samples with predicted bounding boxes and keypoints.rYrOrPrVrQr)r] val_batchz _labels.jpgrypathsfnamenamesrN)r rgrrr)r!r-nis r&plot_val_sampleszPoseValidator.plot_val_samplessdE%L+&%L((,(O{+ *--IbT*EE** LL *r'c Xtj|Dcgc](}|ddddfjdg|j*c}d}t |dgt ||j j||d|jd|d z |j|jd ycc}w) z!Plots predictions for YOLO model.Nr\rVrrY)r;r]rz _pred.jpgr) r`rqrmrr r rr;rrr)r!r-rBrprs r&plot_predictionszPoseValidator.plot_predictionssII5Q5a}qABx}}RA$..A5QSTU E%L *%eTYY5F5FG *" *--IbT*CC** LL  *Rs-B'c t|j}|jr t|n|}t j |ddddf}|ddddfxx|ddddfdz zcc<t |j|jD]g\}}|jj||jt|d|Dcgc]}t|dc}|ddt|dddiycc}w)z.Converts YOLO predictions to COCO JSON format.NrZrUrXrFr\)image_id category_idrxr)score) rstem isnumericintrrziptolistjdictrf class_mapround) r!rfilenamerrboxrbxs r&rtzPoseValidator.pred_to_jsonsH~"" $ 03t9dmmE!RaR%L) ArrE c!QR%j1n$   5DAq JJ  $#~~c!A$i8./0aq!a0qrUqtQ ) *61sDc D|jjr|jrt|jr|j ddz }|j dz }tjd|d|d tddd l m }dd l m }||fD]}|jrJ|d |t|}|j!t|}t#|||d |||d gD]\} } |jrY|j$j&j(Dcgc] }t+t-|j."c}| j0_| j5| j7| j9| dzdz} | j:dd\||j<j>| dz<||j<j>| < |S|Scc}w#t@$r#} tjBd| Yd} ~ |Sd} ~ wwxYw)z8Evaluates object detection model using COCO JSON format.pathz)annotations/person_keypoints_val2017.jsonzpredictions.jsonz" Evaluating pycocotools mAP using z and z...zpycocotools>=2.0.6r)COCO)COCOevalz file not foundrxr)rZrUNrRzpycocotools unable to run: )"rrsis_cocolenrrHrrinforpycocotools.cocorpycocotools.cocoevalris_filerloadResr^r"datasetim_filesrrrparamsimgIdsevaluate accumulate summarizererkeys Exceptionr ) r!re anno_json pred_jsonrrrannorvievalrwes r& eval_jsonzPoseValidator.eval_jsons 99  4<rs1 =)7NNDH&Hr'