o 0j<@sddlmZddlZddlmZddlmZejedGdd d Z ejGd d d Z ejGd d d Z ddZ ddZ ddgdddfddZddZd ddZejGdddZdS)!)ListN)class_requires_deps) benchmarkzopencv-contrib-pythoncsPeZdZdZddeddffdd Zdedefd d Zd edefd d ZZ S) ResizeVideoaRResizes frames of a video to a specified target size. This class provides functionality to resize each frame of a video to a specified square dimension (height and width are equal). Attributes: target_size (int): The desired size (in pixels) for both the height and width of each frame in the video. target_sizereturnNct||_dS)zInitializes the ResizeVideo with a target size. Args: target_size (int): The desired size in pixels for the output frames. Defaults to 224. N)super__init__r )selfr  __class__t/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddlex/inference/models/video_detection/processors.pyr %  zResizeVideo.__init__videoc Csddl}t|}t|d}t|D]1}t|D]*}|||}t|tjr-|j\}} } ntd|j||j |j f|j d|||<qq|S)aResizes all frames of a single video. Args: video (list): A list of segments, where each segment is a list of frames represented as numpy arrays. Returns: list: The input video with each frame resized to the target size. Raises: NotImplementedError: If a frame is not an instance of numpy.ndarray. rNz3Currently, only numpy.ndarray frames are supported.) interpolation) cv2lenrange isinstancenpndarrayshapeNotImplementedErrorresizer Z INTER_LINEAR) rrrnum_segZseg_lenijimghw_rrrr/s$        zResizeVideo.resizevideoscfdd|DS)a7Resizes frames of multiple videos. Args: videos (list): A list containing multiple videos, where each video is a list of segments, and each segment is a list of frames. Returns: list: A list of videos with each frame resized to the target size. cg|]}|qSr)r.0rrrr [z(ResizeVideo.__call__..rrr&rr+r__call__Qs zResizeVideo.__call__)r) __name__ __module__ __qualname____doc__intr rrr/ __classcell__rrrrrs   "rcs`eZdZdZddeddffdd Zdedefd d Zd eeej deej fd d Z Z S) Image2ArrayzJConvert a sequence of images to a numpy array with optional transposition.tchw data_formatr Ncs*t|dvsJd|||_dS)z Initializes the Image2Array class. Args: data_format (str): The format to transpose to, either 'tchw' or 'cthw'. Raises: AssertionError: If data_format is not one of the allowed values. )r7Zcthwz3Target format must be in ['tchw', 'cthw'], but got N)r r r8)rr8rrrr bs  zImage2Array.__init__rcCsPt|}t|D]}||}dd|D}tjdd|Ddd}|||<q|S)a Converts a list of video frames to a numpy array, with frames transposed. Args: video (List): A list of frames represented as numpy arrays. Returns: List: A numpy array with the video frames transposed and concatenated. cSsg|] }|gdqS))r) transposer*r"rrrr,z)Image2Array.img2array..cSsg|] }tj|ddqS)r:axis)r expand_dimsr<rrrr,r=r:r>)rrrZ concatenate)rrrr Z video_onerrr img2arrayss  zImage2Array.img2arrayr&cr')aY Process videos by converting each video to a transposed numpy array. Args: videos (List[List[np.ndarray]]): A list of videos, where each video is a list of frames represented as numpy arrays. Returns: List[np.ndarray]: A list of processed videos with transposed frames. cr(r)rAr)r+rrr,r-z(Image2Array.__call__..rr.rr+rr/ zImage2Array.__call__)r7) r0r1r2r3strr rrArrr/r5rrrrr6^s *r6cspeZdZdZddeddffdd Zdeejdeejfd d Z d eeejdeeejfd d Z Z S)NormalizeVideozH A class to normalize video frames by scaling the pixel values. o@scaler Ncr )z Initializes the NormalizeVideo class. Args: scale (float): The scale factor to normalize the frames, usually the max pixel value. N)r r rF)rrFrrrr rzNormalizeVideo.__init__rcCsJt|}t|D]}||tj|j||<tj||dd||<q|S)a0 Normalizes a sequence of images by scaling the pixel values. Args: video (List[np.ndarray]): A list of frames, where each frame is a numpy array to be normalized. Returns: List[np.ndarray]: The normalized video frames as a list of numpy arrays. rr>)rrastyperfloat32rFr@)rrrr rrrnormalize_videos  zNormalizeVideo.normalize_videor&cr')a^ Apply normalization to a list of videos. Args: videos (List[List[np.ndarray]]): A list of videos, where each video is a list of frames represented as numpy arrays. Returns: List[List[np.ndarray]]: A list of normalized videos, each represented as a list of normalized frames. cr(r)rIr)r+rrr,r-z+NormalizeVideo.__call__..rr.rr+rr/rBzNormalizeVideo.__call__)rE) r0r1r2r3floatr rrrrIr/r5rrrrrDs  .rDcC|d}|S)NrHrGcpu) gpu_matrixZ float_32_grrr convert2cpu rOcCrK)NZint64rL)rNZint_64_grrrconvert2cpu_longrPrQg{Gzt?) g@M-[?gʾ+?g6qrC?gFh@g_{fI?g̒5U@gq@g}8g@gje/ @gYnݭ@r:c+ Cslddl}t||}|dkr|d}|jd}|jdd||ks'J|jd} |jd} g} ||||d|| | g}||d}||d|||| | g}|d| d| } || | dg} || ||ddg} || ||| | g} |d| d| } || | dg } || ||ddg} || ||| | g} |j }||d| }||d| }| |}||||g}|j || tdgddd }| |}||||g}|j || tdgddd }|||dg}||dd| | g}||||| | g}|||dg}||dd| | g}||||| | g}||d|}||d|}||d }|j |dd|d d }||ddg}|j }| ||}|j|dd }|j|dd }||dg}||dg}| | }||}t|}t|}t|}t|}t|}t|}t|}t|D]w}g}t| D]h}t| D]`}t|D]X} ||| ||| |}!||!}"|r||!}#n||!||!}#|#|kr'||!}$||!}%||!}&||!}'||!}(||!})|$| |%| |&| |'| |"|(|)g}*||*qАqʐq| |q| S)a$ Processes the output of a neural network to extract bounding box predictions. Args: output (Tensor): The output tensor from the neural network. conf_thresh (float): The confidence threshold for filtering predictions. Default is 0.005. num_classes (int): The number of classes for classification. Default is 24. anchors (List[float]): A list of anchor box dimensions used in the model. Default is a list of 10 predefined anchor values. num_anchors (int): The number of anchor boxes used in the model. Default is 5. only_objectness (int): If set to 1, only objectness scores are considered for filtering. Default is 1. Returns: all_box(List[List[float]]): A list of predicted bounding boxes for each image in the batch. rNrr:rSr9)r:rr9Zint32)indexr?rT)Z stop_gradientr>)paddlerdimZ unsqueezerZreshaper;ZlinspaceZtiletnnZSigmoid to_tensorZ index_selectrarrayrGexpZSoftmaxmaxZargmaxrOrQrappend)+outputZ conf_thresh num_classesanchorsZ num_anchorsZonly_objectnessrVZ anchor_stepbatchr#r$ all_boxesZgrid_xZgrid_yZsigmoidZxsZysZanchor_wZanchor_hwshs det_confsZ cls_confssZ cls_max_confsZ cls_max_idsZsz_hwZsz_hwabboxescyZcxr inddet_confZconfZbcxZbcybwZbhZ cls_max_confZ cls_max_idboxrrrget_region_boxess!                    roc Csddl}t|dkr |S|t|g}tt|D] }d||d||<q||}g}tt|D]2}|||}|ddkrf||t|dt|D]}|||} t|| dd|kred| d<qPq4|S)zR Performs non-maximum suppression on the input boxes based on their IoUs. rNr:rF)x1y1x2y2)rVrZzerosrZargsortr^bbox_iou) ri nms_threshrVrfr ZsortIdsZ out_boxesZbox_ir!Zbox_jrrrnms]s$      rsTcCsddl}|rKt|d|d}t|d|d}t|d|d}t|d|d}|d|d}|d|d} |d|d} |d|d} nttt|d|ddt|d|dd}tt|d|ddt|d|dd}tt|d|ddt|d|dd}tt|d|ddt|d|dd}|d}|d} |d} |d} ||} ||} || | }| | | }d}|dks|dkr|dS|| }| | }||}|||}||S)zJ Returns the Intersection over Union (IoU) of two bounding boxes. rNr9r:r@g)rVminr]rJrZ)Zbox1Zbox2rprVZmxZMxZmyZMyZw1Zh1Zw2h2ZuwZuhZcwchZcareaZarea1Zarea2Zuarearrrrqvs<2222    rqcs^eZdZdZgfdeeddffdd Zdeded edefd d Zd edefd dZ Z S)DetVideoPostProcesszQ A class used to perform post-processing on detection results in videos. label_listr Ncr )z Args: labels : List[str] A list of labels or class names associated with the detection results. N)r r labels)rryrrrr s zDetVideoPostProcess.__init__predrr score_threshc Csddl}t|}g}t|D]}||}|D]} g} || } t| t|jd} t| jdD]}| |} t| |} | D]|} tt | d| ddd}tt | d| ddd}tt | d| ddd}tt | d| ddd}t | d }tt| d dD]}t | d d| }||}q||kr| ||||g||jt | d gq.r)rrrrr|rrrr/szDetVideoPostProcess.__call__) r0r1r2r3rrCr rJrr/r5rrrrrxs rx)T)typingrnumpyrZ utils.depsrZutils.benchmarkrZtimeitrr6rDrOrQrorsrqrxrrrrs0   D70   '