o 0j@sfddlmZmZmZmZmZddlZddlmZddl m Z m Z m Z ddl mZddlmZd d lmZd d lmZe d rBddlZeeZeeefZejdd de d GdddeZejGdddeZejGdddeZejGdddZejGdddZ ejGdddZ!ejGdddZ"deededeefd d!Z#d"ed#edefd$d%Z$e d  & 'dGd(ed)eeeeefefd*ed+ed,eeefd-e%f d.d/Z&eje d Gd0d1d1Z'd2ed3ee(d4eeefdefd5d6Z)d2ed3ee(d4eeefdefd7d8Z*dHd9d:Z+d;d<Z,dId?d@Z-dAdBZ.dJdCdDZ/ejGdEdFdFZ0dS)K)ListOptionalSequenceTupleUnionN)ndarray)class_requires_depsfunction_requires_depsis_dep_available) ReadImage) benchmark) Normalize)Resizezopencv-contrib-pythonT)nameZis_read_operationc@s<eZdZdZdeeeeefdeefddZ ddZ dS) r z9Reads images from a list of raw image data or file paths.raw_imgsreturncCsg}|D]`}t}t|tr||d<t|tr8d|vr |d}nd|vr-|d}||d<ntd|||}||\}}||d<||d<|jd|jdg|d<|jd|jdg|d<||q|S) akProcesses the input list of raw image data or file paths and returns a list of dictionaries containing image information. Args: raw_imgs (List[Union[ndarray, str]]): A list of raw image data (numpy ndarrays) or file paths (strings). Returns: List[dict]: A list of dictionaries, each containing image information. Zimg_pathimgz@When raw_img is dict, must have one of keys ['img', 'img_path'].ori_imgrimg_size ori_img_size)dict isinstancestr ValueErrorupdatereadshapeappend)selfrZ out_datasZraw_imgdataZsrc_imgrrr$u/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddlex/inference/models/object_detection/processors.py__call__&s.       zReadImage.__call__cCst|tjr|}|jdkrt|tj}||fSt|trI|j |}|dur.t d||}|jdkrE|j dkr>t dt|tj}||fSt dt|jd)NRGBzImage read Error: r zArray is not 3-dimensional.zReadImage only supports the following types: 1. str, indicating a image file path or a directory containing image files. 2. numpy.ndarray. However, got type: .)rnprformatcv2cvtColorZ COLOR_BGR2RGBrZ _img_readerr Exceptionndim RuntimeError TypeErrortype__name__)r"rrZblobr$r$r%rJs(      zReadImage.readN) r2 __module__ __qualname____doc__rrrrrr&rr$r$r$r%r !s$ $r c@&eZdZdeedeefddZdS)rdatasrcCs|D]C}|d}d|vr|jd|jdg|d<|d}||}||d<|jd|jdg}||d<|d|d|d|dg|d<q|S)a< Args: datas (List[dict]): A list of dictionaries, each containing image data with key 'img'. Returns: List[dict]: A list of dictionaries with updated image data, including resized images, original image sizes, resized image sizes, and scale factors. rrrrr scale_factors)r resize)r"r7r#rrrrr$r$r%r&gs   zResize.__call__Nr2r3r4rrr&r$r$r$r%rerc@r6)rr7rcC |D] }||d|d<q|S)zNormalizes images in a list of dictionaries. Iterates over each dictionary, applies normalization to the 'img' key, and returns the modified list. r)Znormr"r7r#r$r$r%r&szNormalize.__call__Nr:r$r$r$r%rr;rc@s*eZdZdZdeedeefddZdS) ToCHWImagezAConverts images in a list of dictionaries from HWC to CHW format.r7rcCs |D] }|dd|d<q|S)aZConverts the image data in the list of dictionaries from HWC to CHW format in-place. Args: datas (List[dict]): A list of dictionaries, each containing an image tensor in 'img' key with HWC format. Returns: List[dict]: The same list of dictionaries with the image tensors converted to CHW format. r)rrr)Z transposer=r$r$r%r&s zToCHWImage.__call__N)r2r3r4r5rrr&r$r$r$r%r>sr>c @sjeZdZdZddeeefddZej fde e dedej d ej fd d Zde e d ee fd d ZdS)ToBatchz Class for batch processing of data dictionaries. Args: ordered_required_keys (Optional[Tuple[str]]): A tuple of keys that need to be present in the input data dictionaries in a specific order. Nordered_required_keyscCs ||_dSNr@)r"r@r$r$r%__init__s zToBatch.__init__r7keydtypercsdkrfdd|D}tj|ddj|ddSdkr2fd d|D}tj|ddj|ddStjfd d|Dddj|ddS) a  Apply batch processing to a list of data dictionaries. Args: datas (List[dict]): A list of data dictionaries to process. key (str): The key in the data dictionaries to extract and batch. dtype (np.dtype): The desired data type of the output array (default is np.float32). Returns: np.ndarray: A numpy array containing the batched data. Raises: KeyError: If the specified key is not found in any of the data dictionaries. rcsg|] }|dddqS)Nr$.0r#rDr$r% sz!ToBatch.apply..r)ZaxisF)rEcopyr8cs&g|]}|ddgdddqS)?NrF)getrGrIr$r%rJs&csg|]}|qSr$r$rGrIr$r%rJs)r)stackastype)r"r7rDrEZ img_sizesr8r$rIr%applysz ToBatch.applycsfddjDS)Ncsg|]}|qSr$rP)rHrDr7r"r$r%rJsz$ToBatch.__call__..rB)r"r7r$rRr%r&szToBatch.__call__rA)r2r3r4r5rrrrCr)float32rrrErrPrr&r$r$r$r%r?s  r?csneZdZdZgdfdeedeeeefffdd Zde de fd d Z d ee dee fd d Z Z S)DetPadz Pad image to a specified size. Args: size (list[int]): image target size fill_value (list[float]): rgb value of pad area, default (114.0, 114.0, 114.0) )\@rUrUsize fill_valuecs,tt|tr||g}||_||_dSrA)superrCrintrVrW)r"rVrW __class__r$r%rCs   zDetPad.__init__rrcCs|}|jdd\}}|j\}}||kr||kr|Stj||dftjd}|tj|jtjd9}|tj|d|d|ddf<|S)Nrr rEr)r rVr)onesrSarrayrWrO)r"rimim_him_whwZcanvasr$r$r%rPs "z DetPad.applyr7cCr<NrrQr=r$r$r%r&zDetPad.__call__)r2r3r4r5rrYrfloatrCrrPrr& __classcell__r$r$rZr%rTs  " rTcsPeZdZdZd deffdd ZdefddZd ee d ee fd d Z Z S) PadStridezpadding image for model with FPN , instead PadBatch(pad_to_stride, pad_gt) in original config Args: stride (bool): model with FPN need image shape % stride == 0 rstridecst||_dSrA)rXrCcoarsest_stride)r"rirZr$r%rCs  zPadStride.__init__rc Cs|}|j}|dkr |S|j\}}}ttt|||}ttt|||}tj|||ftjd} || ddd|d|f<| S)z Args: im (np.ndarray): image (np.ndarray) Returns: im (np.ndarray): processed image (np.ndarray) rr\N)rjr rYr)ceilrfzerosrS) r"rr_rjZim_cr`raZpad_hZpad_wZ padding_imr$r$r%rPs zPadStride.applyr7rcCr<rdrQr=r$r$r%r&rezPadStride.__call__)r) r2r3r4r5rYrCrrPrrr&rgr$r$rZr%rhs "rhpt angle_radrcCsbt|dksJt|t|}}|d||d|}|d||d|}||g}|S)zRotate a point by an angle. Args: pt (list[float]): 2 dimensional point to be rotated angle_rad (float): rotation angle by radian Returns: list[float]: Rotated point. rrr)lenr)sincos)rmrnZsncsZnew_xZnew_yZ rotated_ptr$r$r% rotate_points rsabcCsNt|dksJt|dksJ||}|tj|d |dgtjd}|S)aTo calculate the affine matrix, three pairs of points are required. This function is used to get the 3rd point, given 2D points a & b. The 3rd point is defined by rotating vector `a - b` by 90 degrees anticlockwise, using b as the rotation center. Args: a (np.ndarray): point(x,y) b (np.ndarray): point(x,y) Returns: np.ndarray: The 3rd point. rrrr\)ror)r^rS)rtru directionZthird_ptr$r$r%_get_3rd_point+s  "rwryFcenter input_sizerot output_sizeshiftinvcCst|dksJt|dksJt|dksJt|ttfs)tj||gtjd}|}t|}|d}|d}|d} tj|d} td|dg| } td|dg} tj dtjd} |||| dd d f<|| ||| dd d f<t | dd d f| dd d f| dd d f<tj dtjd}|d | d g|dd d f<t|d | d g| |dd d f<t |dd d f|dd d f|dd d f<|rt t|t| }|St t| t|}|S) aGet the affine transform matrix, given the center/scale/rot/output_size. Args: center (np.ndarray[2, ]): Center of the bounding box (x, y). input_size (np.ndarray[2, ]): Scale of the bounding box wrt [width, height]. rot (float): Rotation angle (degree). output_size (np.ndarray[2, ]): Size of the destination heatmaps. shift (0-100%): Shift translation ratio wrt the width/height. Default (0., 0.). inv (bool): Option to inverse the affine transform direction. (inv=False: src->dst or inv=True: dst->src) Returns: np.ndarray: The transform matrix. rr\rrryg)r rN?) rorrlistr)r^rSpirsrlrwr+ZgetAffineTransform)rzr{r|r}r~rZ scale_tmpZsrc_wZdst_wZdst_hZrot_radsrc_dirZdst_dirsrcdstZtransr$r$r%get_affine_transform>s4 .&.rcsXeZdZdZ       dfdd Zd efd d Zd eedeefddZ Z S) WarpAffineaLApply warp affine transformation to the image based on the given parameters. Args: keep_res (bool): Whether to keep the original resolution aspect ratio during transformation. pad (int): Padding value used when keep_res is True. input_h (int): Target height for the input image when keep_res is False. input_w (int): Target width for the input image when keep_res is False. scale (float): Scale factor for resizing. shift (float): Shift factor for transformation. down_ratio (int): Downsampling ratio for the output image. F皙?皙?rcs8t||_||_||_||_||_||_||_dSrA) rXrCkeep_respadinput_hinput_wscaler~ down_ratio)r"rrrrrr~rrZr$r%rCs  zWarpAffine.__init__rc Cst|tj}|jdd\}}|jr:||jBd}||jBd}tj||gtjd}tj|d|dgtjd}nt ||d}|j |j }}tj|d|dgtjd}t ||d||g}t |||f}tj||||ftjd} |js||j} ||j} t ||d| | g| S)Nrrr\rLg@r)flags)r+r,Z COLOR_RGB2BGRr rrr)r^rSmaxrrrr9Z warpAffineZ INTER_LINEARr) r"rrbrcrrscZ trans_inputinpZout_hZout_wr$r$r%rPs(  zWarpAffine.applyr7rcCsH|D]}|d}d|vr|jd|jdg|d<||}||d<q|S)Nrrrr)r rP)r"r7r#rrr$r$r%r&s  zWarpAffine.__call__)Frrrrrr) r2r3r4r5rCrrPrrr&rgr$r$rZr%rws" rboxeslabelsrc Csg}|\}}|D]E}|dd\}}} } td|}td|}t|| } t|| } | |ks0| |kr1q|t|d|t|dt|d||| | gdq|S)a Restructure the given bounding boxes and labels based on the image size. Args: boxes (ndarray): A 2D array of bounding boxes with each box represented as [cls_id, score, xmin, ymin, xmax, ymax]. labels (List[str]): A list of class labels corresponding to the class ids. img_size (Tuple[int, int]): A tuple representing the width and height of the image. Returns: Boxes: A list of dictionaries, each containing 'cls_id', 'label', 'score', and 'coordinate' keys. rNrrZcls_idlabelscoreZ coordinate)rminr!rYrf) rrrbox_listrcrbboxxminyminxmaxymaxr$r$r%restructured_boxess$        rcCs g}|\}}|jddksJd|D]p}|dd\}}} } } } } }ttd||}ttd||}ttd| |} ttd| |} ttd| |} ttd| |} ttd| |} ttd||}|t|d|t|dt|d||| | | | | |gdq|S)a Restructure the given rotated bounding boxes and labels based on the image size. Args: boxes (ndarray): A 2D array of rotated bounding boxes with each box represented as [cls_id, score, x1, y1, x2, y2, x3, y3, x4, y4]. labels (List[str]): A list of class labels corresponding to the class ids. img_size (Tuple[int, int]): A tuple representing the width and height of the image. Returns: Boxes: A list of dictionaries, each containing 'cls_id', 'label', 'score', and 'coordinate' keys. r z,The shape of rotated boxes should be [N, 10]rNrr)r rrr!rYrf)rrrrrcrbrx1y1x2y2Zx3Zy3Zx4Zy4r$r$r%restructured_rotated_boxess*   rc Cs|dur|St|trng}|D]Y}|\}}}}}} ||vrc||\} } ||} | |} | | }| | }|| d}|| d}||d}||d}||d}||d}|||||||gq||qt|S|dddf|dddf}|dddf|dddf}||d}||d}|dddf|d}|dddf|d}||d}||d}||d}||d}t|dddf|dddf||||f}|S)a] Expand bounding boxes from (x1, y1, x2, y2) format using an unclipping ratio. Parameters: - boxes: np.ndarray of shape (N, 4), where each row is (x1, y1, x2, y2). - unclip_ratio: tuple of (width_ratio, height_ratio), optional. Returns: - expanded_boxes: np.ndarray of shape (N, 4), where each row is (x1, y1, x2, y2). Nrrr rr)rrr!r)r^Z column_stack)rZ unclip_ratioZexpanded_boxesrZclass_idrrrrrZ width_ratioZ height_ratiowidthheightZnew_wZnew_hZcenter_xZcenter_yZnew_x1Znew_y1Znew_x2Znew_y2widthsZheightsr$r$r% unclip_boxessF                   &rcCs|\}}}}|\}}}} t||} t||} t||} t|| } td| | dtd| | d}||d||d}||d| |d}|t|||}|S)z@Compute the Intersection over Union (IoU) of two bounding boxes.rr)rrrf)box1box2rrrrx1_py1_px2_py2_pZx1_iZy1_iZx2_iZy2_iZ inter_area box1_areaZ box2_area iou_valuer$r$r%iouMs      $r333333?ffffff?cCs|dddf}t|ddd}g}t|dkrn|d}||}|d}|d|dd} |||dd}g} |D]&} || } | d} | dd}t| |}|| krZ|n|}||kre| | q?| }t|dks|S)zcPerform Non-Maximum Suppression (NMS) with different IoU thresholds for same and different classes.NrrFrr)r)argsortror!r)riou_sameiou_diffZscoresindicesZselected_boxescurrentZ current_boxZ current_classZcurrent_coordsZfiltered_indicesirZ box_classZ box_coordsr thresholdr$r$r%nmses0        rcCs|\}}}}}}|\}}}}} } ||||} t||} t||} t|| }t|| }td|| }td|| }||}| dkrF|| nd}|dkS)z'Check if box1 is contained within box2.rg?)rr)rr_rrrrrrrrrZxi1Zyi1Zxi2Zyi2Z inter_widthZ inter_heightZintersect_arearr$r$r% is_containeds    rc Cs$t|}tj|td}tj|td}t|D]w}t|D]p}||kr#q|dur8||d|kr8||d|kr8q|dur{|dur{|dkr]||d|kr]t||||r]d||<d||<|dkrz||d|krzt||||rzd||<d||<qt||||rd||<d||<qq||fS)z,Check containment relationships among boxes.r\Nrlargersmall)ror)rlrYranger) r formula_indexcategory_indexmodencontains_othercontained_by_otherrjr$r$r%check_containments4   rcseZdZdZddeeeddffdd Zdede e e fd e e e fd eed ee e e e e fe fd ee ee fdefd dZ    ddee dee d ee e e fd eed ee e e e e ffd eedeefddZZS)DetPostProcesszSave Result Transform This class is responsible for post-processing detection results, including thresholding, non-maximum suppression (NMS), and restructuring the boxes based on the input type (normal or rotated object detection). Nrrcst||_dS)aInitialize the DetPostProcess class. Args: threshold (float, optional): The threshold to apply to the detection scores. Defaults to 0.5. labels (Optional[List[str]], optional): The list of labels for the detection categories. Defaults to None. layout_postprocess (bool, optional): Whether to apply layout post-processing. Defaults to False. N)rXrCr)r"rrZr$r%rCs  zDetPostProcess.__init__rrr layout_nmslayout_unclip_ratiolayout_merge_bboxes_modec"Cs,t|tr"|dddf|k|dddfdk@}||ddf}nPt|trrg}t|dddfD]1} ||dddf| k} |t| d} | dddf| k| dddfdk@} || | q4|rmt|nt g}|rt |ddddfddd } t || }d } | rt |dkr|j dd vr|d|dkrd }nd }d|j vr|j dnd}|d|d}g}|D]E}|dd\}}}}}}||krtd|}td|}t|d|}t|d|}||||}|||kr||q||qt |dkr|}t |}|rd|j vr'|j dnd}t|trl|dvs;Jd||dkrAnt|ddddf|\}}|dkr\||dk}n|dkrk||dk|dkB}npt|trtjt |td}|D]X\}}|dvsJd||dkrq|dkrt|ddddf|||d\}}||dkM}q|dkrt|ddddf|||d\}}||dk|dkBM}q||}|jdkrgS|j ddkrt|dddf |dddff} || }!|!ddddf}|j ddkr/t|dddf} || }!|!ddddf}|rgt|tr=||f}n%t|ttfrQt |dksPJdnt|trXn tdt|dt||}|j ddkry t||j |}|S|j ddkr t||j |}|S td|j d)a Apply post-processing to the detection boxes. Args: boxes (ndarray): The input detection boxes with scores. img_size (tuple): The original image size. Returns: Boxes: The post-processed detection boxes. NrrrFrrg\(\?)rrT)rg= ףp=?g(\?imageZformula)unionrrz\The value of `layout_merge_bboxes_mode` must be one of ['union', 'large', 'small'], but got rrrr\)rrrrz0The length of `layout_unclip_ratio` should be 2.zqThe type of `layout_unclip_ratio` must be float, Tuple[float, float] or Dict[int, Tuple[float, float]], but got r(rz1The shape of boxes should be 6 or 10, instead of ) rrfrr)uniquerMrYr!Zvstackr^rror rindexrrrrr]boolitemsrVZlexsortrtuplerrr1rrr)"r"rrrrrrZ expect_boxesZcategory_filtered_boxesZcat_idZcategory_boxesZcategory_thresholdZselected_indicesZfilter_large_imageZ area_thresZ image_indexZimg_areaZfiltered_boxesrZ label_indexrrrrrZbox_arearrrZ keep_maskrZ layout_modeZ sorted_idxZ sorted_boxesr$r$r%rPs (  $                    (     zDetPostProcess.apply batch_outputsr7c CsBg}t||D]\}} || d|d||||} || q|S)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[Boxes]: The list of post-processed detection boxes. rr)ziprPr!) r"rr7rrrrZoutputsr#outputrr$r$r%r&rs zDetPostProcess.__call__rA)NNNN)r2r3r4r5rrrrCrrrYrrfrrBoxesrPr&rgr$r$rZr%rsL     /r)rxFrA)rr)NNN)1typingrrrrrnumpyr)rZ utils.depsr r r Z common.readerr ZCommonReadImageZutils.benchmarkrcommonrZCommonNormalizerZ CommonResizer+rrrYrfNumberZtimeit_with_optionsZtimeitr>r?rTrhrsrwrrrrrrrrrrrrr$r$r$r%s       B /%"  8M  %  (: "