o 0j@sddlmZmZmZmZmZmZddlZddl m Z ddl m Z m Z ddlmZddlmZd d lmZd d lmZeeefZejGd d d eZe dGdddeZdS))AnyDictListOptionalTupleUnionN)pipeline_requires_extra) HPIConfigPaddlePredictorOption) KptResult) benchmark)(AutoParallelImageSimpleInferencePipeline) BasePipelinecseZdZdZddddddddedeedeedeeeefd eed e d ee eeefe fd dffd dZ dde e ejfded eejejffddZ dde eeeejeejfdeed efddZZS)_KeypointDetectionPipelinezKeypoint Detection pipelineNFdeviceengine engine_config pp_optionuse_hpip hpi_configconfigrrrrrrreturnc stjd ||||||d||dd} i} d|_d| vr*| d| d<| d|_d| vr4| d| d<|j| fi| |_|dd} i} d| vrP| d| d<d | vrZ| d | d <|j| fi| |_|jj|_dS) aInitializes the keypoint detection pipeline. Args: config (Dict): Configuration dictionary containing model and other parameters. device (Optional[str], optional): The device to use for prediction. Defaults to `None`. engine (Optional[str], optional): Inference engine. Defaults to `None`. engine_config (Optional[Dict[str, Any]], optional): Engine-specific config. Defaults to `None`. pp_option (Optional[PaddlePredictorOption], optional): Paddle predictor options. Defaults to `None`. use_hpip (bool, optional): Whether to use HPIP. Defaults to `False`. hpi_config (Optional[Union[Dict[str, Any], HPIConfig]], optional): HPIP configuration. Defaults to `None`. r SubModulesObjectDetectionN thresholdZimgszZKeypointDetectionflipZuse_udp)super__init__ det_thresholdZ create_model det_model kpt_modelZ input_sizekpt_input_size) selfrrrrrrrkwargsZ model_cfgZ model_kwargs __class__r x/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddlex/inference/pipelines/keypoint_detection/pipeline.pyr"!s6       z#_KeypointDetectionPipeline.__init__?bboxpaddingc Cs|dd\}}}}t||||gd}|jd|jd}||||} } | || kr5| |} n | || kr?| |} t| | g|} || fS)a} Convert bounding box from (x1, y1, x2, y2) to center and scale. Args: bbox (Union[Number, np.ndarray]): The bounding box coordinates (x1, y1, x2, y2). padding (float): The padding factor to adjust the scale of the bounding box. Returns: Tuple[np.ndarray, np.ndarray]: The center and scale of the bounding box. Nrg?r)nparrayr&) r'r-r.x1y1Zx2y2centerZ aspect_ratiowhscaler r r+ _box_xyxy2csZs    z'_KeypointDetectionPipeline._box_xyxy2csinputr#c ks|dur|jn|}|j||dD]O}|d|d}}||gd}|dD]4}||d\} } t||| | dd } |d|d|d | d d d | d d d dq&t|VqdS)aPredicts image classification results for the given input. Args: input (str | list[str] | np.ndarray | list[np.ndarray]): The input image(s) or path(s) to the images. det_threshold (float): The detection threshold. Defaults to `None`. **kwargs: Additional keyword arguments that can be passed to the function. Returns: KptResult: The predicted KeyPoint Detection results. N)r input_img input_path)r<r;boxesr= coordinate)Zimgr5r8rZscoreZkpts keypoints kpt_score)r>Z det_scorer?r@)r#r$r9listr%appendr ) r'r:r#r(Zdet_resZori_imgZimg_pathZsingle_img_resboxr5r8Zkpt_resr r r+predictvs4    z"_KeypointDetectionPipeline.predict)r,N)__name__ __module__ __qualname____doc__rrstrrr boolrr r"Numberr0Zndarrayfloatrr9rr rD __classcell__r r r)r+rsR  :  rZcvc@s$eZdZdZeddZddZdS)KeypointDetectionPipelineZhuman_keypoint_detectioncCstSrE)r)r'r r r+ _pipeline_clssz'KeypointDetectionPipeline._pipeline_clscCs|ddddS)NrrZ batch_sizer/)get)r'rr r r+_get_batch_sizesz)KeypointDetectionPipeline._get_batch_sizeN)rFrGrHentitiespropertyrPrRr r r r+rOs   rO)typingrrrrrrnumpyr0Z utils.depsr modelsr r Z models.keypoint_detection.resultr Zutils.benchmarkrZ _parallelrbaserintrMrLZ time_methodsrrOr r r r+s