jO ZddlZddlmZddlmZddlmZgdZdedede e e e e ffd Z Gd d Z d ej d ej dej fdZdej dej dej fdZddej dedeefdZGddZGddZGddZdS)N)anyascii)linear_sum_assignmentPolygon) TextMatchbox_iou polygon_iounmsLocalizationConfusion OCRMetricDetectionMetricword1word2returnc6||k}||k}t|t|k}t|t|k}||||fS)aWPerforms string comparison with multiple levels of tolerance Args: word1: a string word2: another string Returns: a tuple with booleans specifying respectively whether the raw strings, their lower-case counterparts, their anyascii counterparts and their lower-case anyascii counterparts match )lowerr)rr raw_matchcaseless_matchanyascii_match unicase_matchs V/var/www/html/Carbon-Document/venv/lib/python3.11/site-packages/doctr/utils/metrics.py string_matchrs|I[[]]ekkmm3Ne__7NUOO))++x/D/D/F/FFM nnm CCcjeZdZdZd dZdeedeeddfdZdeee ffdZ d d Z dS) ra;Implements text match metric (word-level accuracy) for recognition task. The raw aggregated metric is computed as follows: .. math:: \forall X, Y \in \mathcal{W}^N, TextMatch(X, Y) = \frac{1}{N} \sum\limits_{i=1}^N f_{Y_i}(X_i) with the indicator function :math:`f_{a}` defined as: .. math:: \forall a, x \in \mathcal{W}, f_a(x) = \left\{ \begin{array}{ll} 1 & \mbox{if } x = a \\ 0 & \mbox{otherwise.} \end{array} \right. where :math:`\mathcal{W}` is the set of all possible character sequences, :math:`N` is a strictly positive integer. >>> from doctr.utils import TextMatch >>> metric = TextMatch() >>> metric.update(['Hello', 'world'], ['hello', 'world']) >>> metric.summary() rNc.|dSN)resetselfs r__init__zTextMatch.__init__Is rgtpredct|t|krtdt||D]\}}t||\}}}}|xjt |z c_|xjt |z c_|xjt |z c_|xjt |z c_|xj t|z c_ dS)zUpdate the state of the metric with new predictions Args: gt: list of groung-truth character sequences pred: list of predicted character sequences z(AALP& Nrpolys_1polys_2c|jdks |jdkrtdtj|jd|jdftj}d|D}d|D}t |D]P\}}t |D];\}}||j} |j|jz| z } | | z |||f<zpolygon_iou..999wt}}999rc,g|]}t|Sr@rrhs rrkzpolygon_iou..rlr) ndimr%rJrKrLrM enumerateraarea) rcrdrTshapely_polys_1shapely_polys_2ipoly1jpoly2intersection_area union_areas rr r s|qGLA--FGGGh a('-*:;2:NNNG99999O99999Oo..;;5!/22 ; ;HAu % 2 25 9 9 > ej03DDJ- :GAqDMM ; Nr?boxesthreshc~|dddf}|dddf}|dddf}|dddf}|dddf}||z ||z z}|ddd}g} |jdkrJ|d} | | tj|| ||dd} tj|| ||dd} tj|| ||dd} tj|| ||dd}tjd| | z }tjd|| z }||z}||| ||ddz|z z }tj||kd}||dz}|jdkJ| S) a(Perform non-max suppression, borrowed from `_. Args: boxes: np array of straight boxes: (*, 5), (xmin, ymin, xmax, ymax, score) thresh: iou threshold to perform box suppression. Returns: A list of box indexes to keep NrrGrfrF)argsortsizeappendrJrOrQwhere)rzr{x1y1x2y2scoresareasorderkeeprsxx1yy1xx2yy2whinterovrindss rr r s qqq!tB qqq!tB qqq!tB qqq!tB 111a4[F "Wb !E NN  TTrT "E D *q.. !H AjA59 ..jA59 ..jA59 ..jA59 .. JsC#I & & JsC#I & &AuQx%abb "22U:;xv &&q)dQh *q.. KrceZdZdZ ddededdfdZd ejd ejddfd Z de edzedzedzffd Z dd Z dS)r a Implements common confusion metrics and mean IoU for localization evaluation. The aggregated metrics are computed as follows: .. math:: \forall Y \in \mathcal{B}^N, \forall X \in \mathcal{B}^M, \\ Recall(X, Y) = \frac{1}{N} \sum\limits_{i=1}^N g_{X}(Y_i) \\ Precision(X, Y) = \frac{1}{M} \sum\limits_{i=1}^M g_{X}(Y_i) \\ meanIoU(X, Y) = \frac{1}{M} \sum\limits_{i=1}^M \max\limits_{j \in [1, N]} IoU(X_i, Y_j) with the function :math:`IoU(x, y)` being the Intersection over Union between bounding boxes :math:`x` and :math:`y`, and the function :math:`g_{X}` defined as: .. math:: \forall y \in \mathcal{B}, g_X(y) = \left\{ \begin{array}{ll} 1 & \mbox{if } y\mbox{ has been assigned to any }(X_i)_i\mbox{ with an }IoU \geq 0.5 \\ 0 & \mbox{otherwise.} \end{array} \right. where :math:`\mathcal{B}` is the set of possible bounding boxes, :math:`N` (number of ground truths) and :math:`M` (number of predictions) are strictly positive integers. >>> import numpy as np >>> from doctr.utils import LocalizationConfusion >>> metric = LocalizationConfusion(iou_thresh=0.5) >>> metric.update(np.asarray([[0, 0, 100, 100]]), np.asarray([[0, 0, 70, 70], [110, 95, 200, 150]])) >>> metric.summary() Args: iou_thresh: minimum IoU to consider a pair of prediction and ground truth as a match use_polygons: if set to True, predictions and targets will be expected to have rotated format ryF iou_thresh use_polygonsrNcJ||_||_|dSrrrrrrrs rr zLocalizationConfusion.__init__$ %( rgtspredsc|jddkr|jrt||}nt||}|xjt |dz c_t| \}}|xj t|||f|j kz c_ |xj |jdz c_ |xj |jdz c_ dS)aUpdates the metric Args: gts: a set of relative bounding boxes either of shape (N, 4) or (N, 5) if they are rotated ones preds: a set of relative bounding boxes either of shape (M, 4) or (M, 5) if they are rotated ones rrHN)rLrr rtot_iour?maxsumrmatchesr(rnum_gts num_preds)rrrrT gt_indices pred_indicess rr2zLocalizationConfusion.update s ;q>A    .%c511!#u-- LLE'++1+"5"5"9"9";";<< 11T ?Cnq>P>P56:::VZy(**rc>d|_d|_d|_d|_dSNrr)rrrrrs rrzLocalizationConfusion.reset1s"   rryFr8 r9r:r;r<r?boolr rJndarrayr2tupler6rr@rrr r s""L "  )"*)RZ)D)))).+ut|UT\54<GH++++"rr c eZdZdZ ddededdfdZd ejd ejd e e d e e ddf d Z de e e edzfe e edzfedzffdZddZdS)r a\Implements an end-to-end OCR metric. The aggregated metrics are computed as follows: .. math:: \forall (B, L) \in \mathcal{B}^N \times \mathcal{L}^N, \forall (\hat{B}, \hat{L}) \in \mathcal{B}^M \times \mathcal{L}^M, \\ Recall(B, \hat{B}, L, \hat{L}) = \frac{1}{N} \sum\limits_{i=1}^N h_{B,L}(\hat{B}_i, \hat{L}_i) \\ Precision(B, \hat{B}, L, \hat{L}) = \frac{1}{M} \sum\limits_{i=1}^M h_{B,L}(\hat{B}_i, \hat{L}_i) \\ meanIoU(B, \hat{B}) = \frac{1}{M} \sum\limits_{i=1}^M \max\limits_{j \in [1, N]} IoU(\hat{B}_i, B_j) with the function :math:`IoU(x, y)` being the Intersection over Union between bounding boxes :math:`x` and :math:`y`, and the function :math:`h_{B, L}` defined as: .. math:: \forall (b, l) \in \mathcal{B} \times \mathcal{L}, h_{B,L}(b, l) = \left\{ \begin{array}{ll} 1 & \mbox{if } b\mbox{ has been assigned to a given }B_j\mbox{ with an } \\ & IoU \geq 0.5 \mbox{ and that for this assignment, } l = L_j\\ 0 & \mbox{otherwise.} \end{array} \right. where :math:`\mathcal{B}` is the set of possible bounding boxes, :math:`\mathcal{L}` is the set of possible character sequences, :math:`N` (number of ground truths) and :math:`M` (number of predictions) are strictly positive integers. >>> import numpy as np >>> from doctr.utils import OCRMetric >>> metric = OCRMetric(iou_thresh=0.5) >>> metric.update(np.asarray([[0, 0, 100, 100]]), np.asarray([[0, 0, 70, 70], [110, 95, 200, 150]]), >>> ['hello'], ['hello', 'world']) >>> metric.summary() Args: iou_thresh: minimum IoU to consider a pair of prediction and ground truth as a match use_polygons: if set to True, predictions and targets will be expected to have rotated format ryFrrrNcJ||_||_|dSrrrs rr zOCRMetric.__init__arrgt_boxes pred_boxes gt_labels pred_labelsc|jdt|ks|jdt|krtd|jddkrH|jrt ||}nt ||}|xjt|d z c_t| \}}|||f|j k}t||||D]\} } t|| || \} } } }|xjt| z c_|xjt| z c_|xjt| z c_|xjt|z c_|xj|jdz c_|xj|jdz c_dS)arUpdates the metric Args: gt_boxes: a set of relative bounding boxes either of shape (N, 4) or (N, 5) if they are rotated ones pred_boxes: a set of relative bounding boxes either of shape (M, 4) or (M, 5) if they are rotated ones gt_labels: a list of N string labels pred_labels: a list of M string labels rathere should be the same number of boxes and string both for the ground truth and the predictionsrHN)rLr$r%rr rrr?rrrrr&r raw_matchesr(caseless_matchesanyascii_matchesunicase_matchesrr)rrrrrrTrris_keptgt_idxpred_idxr.r/r0r1s rr2zOCRMetric.updatejs >! I . .*2B12E[IYIY2Y2Y s   A  " "  8%h ;;!(J77 LLE'++1+"5"5"9"9";";<< 11T?C~PQ?Q?QT*T^;;W[?C~PQ?Q?QT*T^;;W[=A^a=O=OD(4>99UY    ?Cnq>P>P56:::VZy(**rchd|_d|_d|_d|_d|_d|_d|_dSr)rrrrrrrrs rrzOCRMetric.resets=   ! ! rrr8)r9r:r;r<r?rr rJrr=r>r2rr5r6rr@rrr r 8s &&T "  ).*).J).9 ). #Y ).  ).).).).V+tC$56S%$,=N8OQVY]Q]]^++++6!!!!!!rr c eZdZdZ ddededdfdZd ejd ejd ejd ejddf d Z de edzedzedzffdZ ddZ dS)r aImplements an object detection metric. The aggregated metrics are computed as follows: .. math:: \forall (B, C) \in \mathcal{B}^N \times \mathcal{C}^N, \forall (\hat{B}, \hat{C}) \in \mathcal{B}^M \times \mathcal{C}^M, \\ Recall(B, \hat{B}, C, \hat{C}) = \frac{1}{N} \sum\limits_{i=1}^N h_{B,C}(\hat{B}_i, \hat{C}_i) \\ Precision(B, \hat{B}, C, \hat{C}) = \frac{1}{M} \sum\limits_{i=1}^M h_{B,C}(\hat{B}_i, \hat{C}_i) \\ meanIoU(B, \hat{B}) = \frac{1}{M} \sum\limits_{i=1}^M \max\limits_{j \in [1, N]} IoU(\hat{B}_i, B_j) with the function :math:`IoU(x, y)` being the Intersection over Union between bounding boxes :math:`x` and :math:`y`, and the function :math:`h_{B, C}` defined as: .. math:: \forall (b, c) \in \mathcal{B} \times \mathcal{C}, h_{B,C}(b, c) = \left\{ \begin{array}{ll} 1 & \mbox{if } b\mbox{ has been assigned to a given }B_j\mbox{ with an } \\ & IoU \geq 0.5 \mbox{ and that for this assignment, } c = C_j\\ 0 & \mbox{otherwise.} \end{array} \right. where :math:`\mathcal{B}` is the set of possible bounding boxes, :math:`\mathcal{C}` is the set of possible class indices, :math:`N` (number of ground truths) and :math:`M` (number of predictions) are strictly positive integers. >>> import numpy as np >>> from doctr.utils import DetectionMetric >>> metric = DetectionMetric(iou_thresh=0.5) >>> metric.update(np.asarray([[0, 0, 100, 100]]), np.asarray([[0, 0, 70, 70], [110, 95, 200, 150]]), >>> np.zeros(1, dtype=np.int64), np.array([0, 1], dtype=np.int64)) >>> metric.summary() Args: iou_thresh: minimum IoU to consider a pair of prediction and ground truth as a match use_polygons: if set to True, predictions and targets will be expected to have rotated format ryFrrrNcJ||_||_|dSrrrs rr zDetectionMetric.__init__rrrrrrc|jd|jdks|jd|jdkrtd|jddkr|jrt||}nt ||}|xjt |dz c_t| \}}|||f|j k}|xj t||||||kz c_ |xj |jdz c_ |xj|jdz c_dS)aUpdates the metric Args: gt_boxes: a set of relative bounding boxes either of shape (N, 4) or (N, 5) if they are rotated ones pred_boxes: a set of relative bounding boxes either of shape (M, 4) or (M, 5) if they are rotated ones gt_labels: an array of class indices of shape (N,) pred_labels: an array of class indices of shape (M,) rrrHN)rLr%rr rrr?rrrr num_matchesr(rr) rrrrrrTrrrs rr2zDetectionMetric.updatesa >!   2 2 2j6Fq6I[M^_`Ma6a6a s   A  " "  8%h ;;!(J77 LLE'++1+"5"5"9"9";";<< !9K9KD$t~55QU ?Cnq>P>P56:::VZy(**rc>d|_d|_d|_d|_dSr)rrrrrs rrzDetectionMetric.reset#s%  rrr8rr@rrr r s&&T "  $.*$.J$.: $. Z $.  $.$.$.$.L+ut|UT\54<GH++++"rr )ry)numpyrJrscipy.optimizershapely.geometryr__all__r>rrrrrrr r?r=r(r r r r r@rrrs000000$$$$$$   DDCDE$dD2H,IDDDD*MMMMMMMM`RZ"*8bjRZ:##rz#5#49####LZZZZZZZZz!!!!!!!!Dmmmmmmmmmmr