o 0j6 @s&ddlZddlmZmZmZmZmZddlZddlm Z ddl m Z m Z ddl mZdd lmZe d r8ddlZeeefZeeZd ed e d e de de f ddZeje d GdddZde de fddZde deeefdeeefdeeefde f ddZeje d GdddZdS) N)ListOptionalSequenceTupleUnion)ndarray)class_requires_depsis_dep_available) benchmark)get_affine_transformzopencv-contrib-pythontheta size_inputsize_dst size_targetreturncCst|}tjdtjd}|d|d}|d|d}t|||d<t| ||d<|d|dt|d|dt|d|d|d <t|||d <t|||d <|d|dt|d|dt|d|d|d <|S) aThis code is based on https://github.com/open-mmlab/mmpose/blob/master/mmpose/core/post_processing/post_transforms.py Calculate the transformation matrix under the constraint of unbiased. Paper ref: Huang et al. The Devil is in the Details: Delving into Unbiased Data Processing for Human Pose Estimation (CVPR 2020). Args: theta (float): Rotation angle in degrees. size_input (np.ndarray): Size of input image [w, h]. size_dst (np.ndarray): Size of output image [w, h]. size_target (np.ndarray): Size of ROI in input plane [w, h]. Returns: matrix (np.ndarray): A matrix for transformation. )r r )Zdtyper)rr)rrg?)rr )rr)rr)rr )npZdeg2radzerosfloat32cossin)rrrrmatrixZscale_xZscale_yrw/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddlex/inference/models/keypoint_detection/processors.pyget_warp_matrix s.   rc @seZdZdZddeeefdefddZ  dded e e ee e fefd e e ee e fefd eeeeffd d Z de ed e efddZdS) TopDownAffinea:refer to https://github.com/open-mmlab/mmpose/blob/71ec36ebd63c475ab589afc817868e749a61491f/mmpose/datasets/transforms/topdown_transforms.py#L13 Get the bbox image as the model input by affine transform. Args: input_size (Tuple[int, int]): The input image size of the model in [w, h]. The bbox region will be cropped and resize to `input_size` use_udp (bool): Whether use unbiased data processing. See `UDP (CVPR 2020)`_ for details. Defaults to ``False`` .. _`UDP (CVPR 2020)`: https://arxiv.org/abs/1911.07524 F input_sizeuse_udpcCs<tdd|Drt|dksJd|||_||_dS)NcSsg|]}t|tqSr) isinstanceint).0irrr Ysz*TopDownAffine.__init__..r zInvalid input_size )alllenr r!)selfr r!rrr__init__Ws zTopDownAffine.__init__NimgcenterscalercCsd}t|jddddd}t|trt|}t|tr%t|}|dur+|n|d}|dur5|n|}|jrft||d|jdd|jddg|}tj ||t |jdt |jdftj d}nt ||||j}tj ||t |jdt |jdftj d}|||fS) aVApplies a wrapaffine to the input image based on the specified center, scale. Args: img (ndarray): The input image as a NumPy ndarray. center (Optional[Union[Tuple[Number, Number], ndarray]], optional): Center of the bounding box (x, y) scale (Optional[Union[Tuple[Number, Number], ndarray]], optional): Scale of the bounding box wrt [width, height]. Returns: Tuple[ndarray, ndarray, ndarray]: The transformed image, the center used for the transformation, and the scale used for the transformation. rNr g@?r)flags) rarrayshaper"rr!rr cv2Z warpAffiner#Z INTER_LINEARr)r)r+r,r-ZrotZimshapetransrrrapply^s:     zTopDownAffine.applydatasc Cs|D]J}|d}d|vr||d<d|vr |jd|jdg|d<|||dd|dd\}}}||d<||d<||d<|jd|jdg}||d<q|S) Nr+ori_imgZ ori_img_sizerrr,r-img_size)r2r5get)r)r6datar7r+r,r-r8rrr__call__s  zTopDownAffine.__call__)F)NN)__name__ __module__ __qualname____doc__rr#boolr*rrrNumberr5rdictr;rrrrrHs    3rpttcCs2t|d|ddgj}t||}|ddS)aApply an affine transformation to a 2D point. Args: pt (numpy.ndarray): A 2D point represented as a 2-element array. t (numpy.ndarray): A 3x3 affine transformation matrix. Returns: numpy.ndarray: The transformed 2D point. rrr/Nr )rr1Tdot)rCrDZnew_ptrrraffine_transforms  rGcoordsr,r- output_sizecCsXt|j}t||d|dd}t|jdD]}t||ddf|||ddf<q|S)aTransform coordinates to the target space using an affine transformation. Args: coords (numpy.ndarray): Original coordinates, shape (N, 2). center (tuple): Center point for the transformation. scale (tuple): Scale factor for the transformation. output_size (tuple): Size of the output space. Returns: numpy.ndarray: Transformed coordinates, shape (N, 2). rr)invr )rrr2rrangerG)rHr,r-rIZ target_coordsr4prrrtransform_predss $rMc @seZdZdZd!ddZdedededefd d Zd ee d ee deefd dZ d"dededede fddZ dede eeffddZdede defddZdedefddZdedede defddZd S)#KptPostProcesszSave Result TransformTcCs ||_dS)N)use_dark)r)rOrrrr*s zKptPostProcess.__init__heatmapr,r-rcCsj|d|d|d}}}||||\}}tj||fddtj|ddd}}ddt||DS)r5)N.r.ZaxisrcSsg|] \}}||dqS)) keypointsZ kpt_scorer)r$ZkptZscorerrrr&sz(KptPostProcess.apply..)get_final_predsrZ concatenatemeansqueezezip)r)rPr,r-predsmaxvalsrRZscoresrrrr5s  zKptPostProcess.apply batch_outputsr6csfddt||DS)aApply the post-processing to a batch of outputs. Args: batch_outputs (List[dict]): The list of detection outputs. datas (List[dict]): The list of input data. Returns: List[dict]: The list of post-processed keypoints. cs*g|]\}}|d|d|dqS)rPr,r-)r5)r$r:outputr)rrr&sz+KptPostProcess.__call__..)rV)r)rYr6rr[rr;s zKptPostProcess.__call__r heatmaps kernelsizec Cs||\}}|jd}|jd}|jr||||}nt|jdD]~} t|jdD]t} || | } tt|| | dd} tt|| | dd} d| kr^|dkrnq,d| krl|dkrnq,t | | | d| | | d| | d| | | d| g}|| | t |d7<q,q#| }t|jdD]}t ||||||||g||<q||fS)athe highest heatvalue location with a quarter offset in the direction from the highest response to the second highest response. Args: heatmaps (numpy.ndarray): The predicted heatmaps center (numpy.ndarray): The boxes center scale (numpy.ndarray): The scale factor Returns: preds: numpy.ndarray([batch_size, num_joints, 2]), keypoints coords maxvals: numpy.ndarray([batch_size, num_joints, 1]), the maximum confidence of the keypoints r r rrr?) get_max_predsr2rOdark_postprocessrKr#mathfloorrr1signcopyrM)r)r\r,r-r]rHrXheatmap_height heatmap_widthnrLhmpxpydiffrWr%rrrrSs2   8  zKptPostProcess.get_final_predsc Cs.t|tjs Jd|jdksJd|jd}|jd}|jd}|||df}t|d}t|d}|||df}|||df}t|d  tj }|d d d d df||d d d d df<t |d d d d df||d d d d df<tt |d d } | tj } || 9}||fS) aMget predictions from score maps Args: heatmaps: numpy.ndarray([batch_size, num_joints, height, width]) Returns: preds: numpy.ndarray([batch_size, num_joints, 2]), keypoints coords maxvals: numpy.ndarray([batch_size, num_joints, 2]), the maximum confidence of the keypoints z heatmaps should be numpy.ndarrayrzbatch_images should be 4-ndimrrr r.r )rrr Ng) r"rrndimr2ZreshapeZargmaxZamaxZtileZastyperrbZgreater) r)r\ batch_size num_jointswidthZheatmaps_reshapedidxrXrWZ pred_maskrrrr_s"     ,2 zKptPostProcess.get_max_predskernelc Cs|dd}|jd}|jd}|jd}|jd}t|D]a}t|D]Z} t||| f} t|d||d|f} ||| f| || || f<t| ||fd} | || || f||| f<||| f| t||| f9<q$q|S)Nrr rr )r2rKrmaxrrdr3Z GaussianBlur) r)rPrqborderrmrnheightror%jZ origin_maxZdrrrr gaussian_blur7s       $$(zKptPostProcess.gaussian_blurrhcoordcCs|jd}|jd}t|d}t|d}d|kr"|dkrn|Sd|kr1|dkrn|Sd|||d|||d}d||d|||d|}d|||dd||||||d} d||d|d||d|d||d|d||d|d} d||d|d|||||d|} t|g|gg} t| | g| | gg} | | | ddkr| j}| | }tjt|jdd}||7}|S)Nrrr rr^rQ)r2r#rrIrUr1rE)r)rhrwrerfrirjZdxZdyZdxxZdxyZdyyZ derivativeZhessianZ hessianinvoffsetrrr dark_parseHs:    $$44 zKptPostProcess.dark_parserHcCsr|||}t|d}t|}t|jdD]}t|jdD]}||||||||||f<q!q|S)zV refer to https://github.com/ilovepose/DarkPose/lib/core/inference.py g|=rr)rvrmaximumlogrKr2rz)r)rhrHr]rgrLrrrr`bs   &zKptPostProcess.dark_postprocessN)T)r )r<r=r>r?r*rKptsr5rrBr;r#rSrr_rvrzr`rrrrrNs6 " *rN)ratypingrrrrrnumpyrrZ utils.depsr r Zutils.benchmarkr Zobject_detection.processorsrr3r#floatrArBr}rZtimeitrrGrMrNrrrrsP     (\