Deteval.py 34.0 KB
Newer Older
1
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
J
Jethong 已提交
2 3 4 5 6 7 8 9 10 11 12 13
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
14

J
Jethong 已提交
15 16 17 18 19 20 21
import numpy as np
from ppocr.utils.e2e_metric.polygon_fast import iod, area_of_intersection, area


def get_socre(gt_dict, pred_dict):
    allInputs = 1

22
    def input_reading_mod(pred_dict):
J
Jethong 已提交
23 24 25 26 27 28 29 30 31 32
        """This helper reads input from txt files"""
        det = []
        n = len(pred_dict)
        for i in range(n):
            points = pred_dict[i]['points']
            text = pred_dict[i]['text']
            point = ",".join(map(str, points.reshape(-1, )))
            det.append([point, text])
        return det

33
    def gt_reading_mod(gt_dict):
J
Jethong 已提交
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109
        """This helper reads groundtruths from mat files"""
        gt = []
        n = len(gt_dict)
        for i in range(n):
            points = gt_dict[i]['points'].tolist()
            h = len(points)
            text = gt_dict[i]['text']
            xx = [
                np.array(
                    ['x:'], dtype='<U2'), 0, np.array(
                        ['y:'], dtype='<U2'), 0, np.array(
                            ['#'], dtype='<U1'), np.array(
                                ['#'], dtype='<U1')
            ]
            t_x, t_y = [], []
            for j in range(h):
                t_x.append(points[j][0])
                t_y.append(points[j][1])
            xx[1] = np.array([t_x], dtype='int16')
            xx[3] = np.array([t_y], dtype='int16')
            if text != "":
                xx[4] = np.array([text], dtype='U{}'.format(len(text)))
                xx[5] = np.array(['c'], dtype='<U1')
            gt.append(xx)
        return gt

    def detection_filtering(detections, groundtruths, threshold=0.5):
        for gt_id, gt in enumerate(groundtruths):
            if (gt[5] == '#') and (gt[1].shape[1] > 1):
                gt_x = list(map(int, np.squeeze(gt[1])))
                gt_y = list(map(int, np.squeeze(gt[3])))
                for det_id, detection in enumerate(detections):
                    detection_orig = detection
                    detection = [float(x) for x in detection[0].split(',')]
                    detection = list(map(int, detection))
                    det_x = detection[0::2]
                    det_y = detection[1::2]
                    det_gt_iou = iod(det_x, det_y, gt_x, gt_y)
                    if det_gt_iou > threshold:
                        detections[det_id] = []

                detections[:] = [item for item in detections if item != []]
        return detections

    def sigma_calculation(det_x, det_y, gt_x, gt_y):
        """
        sigma = inter_area / gt_area
        """
        return np.round((area_of_intersection(det_x, det_y, gt_x, gt_y) /
                         area(gt_x, gt_y)), 2)

    def tau_calculation(det_x, det_y, gt_x, gt_y):
        if area(det_x, det_y) == 0.0:
            return 0
        return np.round((area_of_intersection(det_x, det_y, gt_x, gt_y) /
                         area(det_x, det_y)), 2)

    ##############################Initialization###################################
    global_tp = 0
    global_fp = 0
    global_fn = 0
    global_sigma = []
    global_tau = []
    tr = 0.7
    tp = 0.6
    fsc_k = 0.8
    k = 2
    global_pred_str = []
    global_gt_str = []
    ###############################################################################

    for input_id in range(allInputs):
        if (input_id != '.DS_Store') and (input_id != 'Pascal_result.txt') and (
                input_id != 'Pascal_result_curved.txt') and (input_id != 'Pascal_result_non_curved.txt') and (
                input_id != 'Deteval_result.txt') and (input_id != 'Deteval_result_curved.txt') \
                and (input_id != 'Deteval_result_non_curved.txt'):
110 111
            detections = input_reading_mod(pred_dict)
            groundtruths = gt_reading_mod(gt_dict)
J
Jethong 已提交
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198
            detections = detection_filtering(
                detections,
                groundtruths)  # filters detections overlapping with DC area
            dc_id = []
            for i in range(len(groundtruths)):
                if groundtruths[i][5] == '#':
                    dc_id.append(i)
            cnt = 0
            for a in dc_id:
                num = a - cnt
                del groundtruths[num]
                cnt += 1

            local_sigma_table = np.zeros((len(groundtruths), len(detections)))
            local_tau_table = np.zeros((len(groundtruths), len(detections)))
            local_pred_str = {}
            local_gt_str = {}

            for gt_id, gt in enumerate(groundtruths):
                if len(detections) > 0:
                    for det_id, detection in enumerate(detections):
                        detection_orig = detection
                        detection = [float(x) for x in detection[0].split(',')]
                        detection = list(map(int, detection))
                        pred_seq_str = detection_orig[1].strip()
                        det_x = detection[0::2]
                        det_y = detection[1::2]
                        gt_x = list(map(int, np.squeeze(gt[1])))
                        gt_y = list(map(int, np.squeeze(gt[3])))
                        gt_seq_str = str(gt[4].tolist()[0])

                        local_sigma_table[gt_id, det_id] = sigma_calculation(
                            det_x, det_y, gt_x, gt_y)
                        local_tau_table[gt_id, det_id] = tau_calculation(
                            det_x, det_y, gt_x, gt_y)
                        local_pred_str[det_id] = pred_seq_str
                        local_gt_str[gt_id] = gt_seq_str

            global_sigma.append(local_sigma_table)
            global_tau.append(local_tau_table)
            global_pred_str.append(local_pred_str)
            global_gt_str.append(local_gt_str)

    global_accumulative_recall = 0
    global_accumulative_precision = 0
    total_num_gt = 0
    total_num_det = 0
    hit_str_count = 0
    hit_count = 0

    def one_to_one(local_sigma_table, local_tau_table,
                   local_accumulative_recall, local_accumulative_precision,
                   global_accumulative_recall, global_accumulative_precision,
                   gt_flag, det_flag, idy):
        hit_str_num = 0
        for gt_id in range(num_gt):
            gt_matching_qualified_sigma_candidates = np.where(
                local_sigma_table[gt_id, :] > tr)
            gt_matching_num_qualified_sigma_candidates = gt_matching_qualified_sigma_candidates[
                0].shape[0]
            gt_matching_qualified_tau_candidates = np.where(
                local_tau_table[gt_id, :] > tp)
            gt_matching_num_qualified_tau_candidates = gt_matching_qualified_tau_candidates[
                0].shape[0]

            det_matching_qualified_sigma_candidates = np.where(
                local_sigma_table[:, gt_matching_qualified_sigma_candidates[0]]
                > tr)
            det_matching_num_qualified_sigma_candidates = det_matching_qualified_sigma_candidates[
                0].shape[0]
            det_matching_qualified_tau_candidates = np.where(
                local_tau_table[:, gt_matching_qualified_tau_candidates[0]] >
                tp)
            det_matching_num_qualified_tau_candidates = det_matching_qualified_tau_candidates[
                0].shape[0]

            if (gt_matching_num_qualified_sigma_candidates == 1) and (gt_matching_num_qualified_tau_candidates == 1) and \
                    (det_matching_num_qualified_sigma_candidates == 1) and (
                    det_matching_num_qualified_tau_candidates == 1):
                global_accumulative_recall = global_accumulative_recall + 1.0
                global_accumulative_precision = global_accumulative_precision + 1.0
                local_accumulative_recall = local_accumulative_recall + 1.0
                local_accumulative_precision = local_accumulative_precision + 1.0

                gt_flag[0, gt_id] = 1
                matched_det_id = np.where(local_sigma_table[gt_id, :] > tr)
                # recg start
199

J
Jethong 已提交
200 201 202
                gt_str_cur = global_gt_str[idy][gt_id]
                pred_str_cur = global_pred_str[idy][matched_det_id[0].tolist()[
                    0]]
203

J
Jethong 已提交
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
                if pred_str_cur == gt_str_cur:
                    hit_str_num += 1
                else:
                    if pred_str_cur.lower() == gt_str_cur.lower():
                        hit_str_num += 1
                # recg end
                det_flag[0, matched_det_id] = 1
        return local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, gt_flag, det_flag, hit_str_num

    def one_to_many(local_sigma_table, local_tau_table,
                    local_accumulative_recall, local_accumulative_precision,
                    global_accumulative_recall, global_accumulative_precision,
                    gt_flag, det_flag, idy):
        hit_str_num = 0
        for gt_id in range(num_gt):
            # skip the following if the groundtruth was matched
            if gt_flag[0, gt_id] > 0:
                continue

            non_zero_in_sigma = np.where(local_sigma_table[gt_id, :] > 0)
            num_non_zero_in_sigma = non_zero_in_sigma[0].shape[0]

            if num_non_zero_in_sigma >= k:
                ####search for all detections that overlaps with this groundtruth
                qualified_tau_candidates = np.where((local_tau_table[
                    gt_id, :] >= tp) & (det_flag[0, :] == 0))
                num_qualified_tau_candidates = qualified_tau_candidates[
                    0].shape[0]

                if num_qualified_tau_candidates == 1:
                    if ((local_tau_table[gt_id, qualified_tau_candidates] >= tp)
                            and
                        (local_sigma_table[gt_id, qualified_tau_candidates] >=
                         tr)):
                        # became an one-to-one case
                        global_accumulative_recall = global_accumulative_recall + 1.0
                        global_accumulative_precision = global_accumulative_precision + 1.0
                        local_accumulative_recall = local_accumulative_recall + 1.0
                        local_accumulative_precision = local_accumulative_precision + 1.0

                        gt_flag[0, gt_id] = 1
                        det_flag[0, qualified_tau_candidates] = 1
                        # recg start
                        gt_str_cur = global_gt_str[idy][gt_id]
                        pred_str_cur = global_pred_str[idy][
                            qualified_tau_candidates[0].tolist()[0]]
250

J
Jethong 已提交
251 252 253 254 255 256 257 258 259 260 261
                        if pred_str_cur == gt_str_cur:
                            hit_str_num += 1
                        else:
                            if pred_str_cur.lower() == gt_str_cur.lower():
                                hit_str_num += 1
                        # recg end
                elif (np.sum(local_sigma_table[gt_id, qualified_tau_candidates])
                      >= tr):
                    gt_flag[0, gt_id] = 1
                    det_flag[0, qualified_tau_candidates] = 1
                    # recg start
262

J
Jethong 已提交
263 264 265
                    gt_str_cur = global_gt_str[idy][gt_id]
                    pred_str_cur = global_pred_str[idy][
                        qualified_tau_candidates[0].tolist()[0]]
266

J
Jethong 已提交
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
                    if pred_str_cur == gt_str_cur:
                        hit_str_num += 1
                    else:
                        if pred_str_cur.lower() == gt_str_cur.lower():
                            hit_str_num += 1
                    # recg end

                    global_accumulative_recall = global_accumulative_recall + fsc_k
                    global_accumulative_precision = global_accumulative_precision + num_qualified_tau_candidates * fsc_k

                    local_accumulative_recall = local_accumulative_recall + fsc_k
                    local_accumulative_precision = local_accumulative_precision + num_qualified_tau_candidates * fsc_k

        return local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, gt_flag, det_flag, hit_str_num

    def many_to_one(local_sigma_table, local_tau_table,
                    local_accumulative_recall, local_accumulative_precision,
                    global_accumulative_recall, global_accumulative_precision,
                    gt_flag, det_flag, idy):
        hit_str_num = 0
        for det_id in range(num_det):
            # skip the following if the detection was matched
            if det_flag[0, det_id] > 0:
                continue

            non_zero_in_tau = np.where(local_tau_table[:, det_id] > 0)
            num_non_zero_in_tau = non_zero_in_tau[0].shape[0]

            if num_non_zero_in_tau >= k:
                ####search for all detections that overlaps with this groundtruth
                qualified_sigma_candidates = np.where((
                    local_sigma_table[:, det_id] >= tp) & (gt_flag[0, :] == 0))
                num_qualified_sigma_candidates = qualified_sigma_candidates[
                    0].shape[0]

                if num_qualified_sigma_candidates == 1:
                    if ((local_tau_table[qualified_sigma_candidates, det_id] >=
                         tp) and
                        (local_sigma_table[qualified_sigma_candidates, det_id]
                         >= tr)):
                        # became an one-to-one case
                        global_accumulative_recall = global_accumulative_recall + 1.0
                        global_accumulative_precision = global_accumulative_precision + 1.0
                        local_accumulative_recall = local_accumulative_recall + 1.0
                        local_accumulative_precision = local_accumulative_precision + 1.0

                        gt_flag[0, qualified_sigma_candidates] = 1
                        det_flag[0, det_id] = 1
                        # recg start
                        pred_str_cur = global_pred_str[idy][det_id]
                        gt_len = len(qualified_sigma_candidates[0])
                        for idx in range(gt_len):
                            ele_gt_id = qualified_sigma_candidates[0].tolist()[
                                idx]
321
                            if ele_gt_id not in global_gt_str[idy]:
J
Jethong 已提交
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
                                continue
                            gt_str_cur = global_gt_str[idy][ele_gt_id]
                            if pred_str_cur == gt_str_cur:
                                hit_str_num += 1
                                break
                            else:
                                if pred_str_cur.lower() == gt_str_cur.lower():
                                    hit_str_num += 1
                                break
                        # recg end
                elif (np.sum(local_tau_table[qualified_sigma_candidates,
                                             det_id]) >= tp):
                    det_flag[0, det_id] = 1
                    gt_flag[0, qualified_sigma_candidates] = 1
                    # recg start
337

J
Jethong 已提交
338 339 340 341
                    pred_str_cur = global_pred_str[idy][det_id]
                    gt_len = len(qualified_sigma_candidates[0])
                    for idx in range(gt_len):
                        ele_gt_id = qualified_sigma_candidates[0].tolist()[idx]
342
                        if ele_gt_id not in global_gt_str[idy]:
J
Jethong 已提交
343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623
                            continue
                        gt_str_cur = global_gt_str[idy][ele_gt_id]
                        if pred_str_cur == gt_str_cur:
                            hit_str_num += 1
                            break
                        else:
                            if pred_str_cur.lower() == gt_str_cur.lower():
                                hit_str_num += 1
                                break
                    # recg end

                    global_accumulative_recall = global_accumulative_recall + num_qualified_sigma_candidates * fsc_k
                    global_accumulative_precision = global_accumulative_precision + fsc_k

                    local_accumulative_recall = local_accumulative_recall + num_qualified_sigma_candidates * fsc_k
                    local_accumulative_precision = local_accumulative_precision + fsc_k
        return local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, gt_flag, det_flag, hit_str_num

    single_data = {}
    for idx in range(len(global_sigma)):
        local_sigma_table = global_sigma[idx]
        local_tau_table = global_tau[idx]

        num_gt = local_sigma_table.shape[0]
        num_det = local_sigma_table.shape[1]

        total_num_gt = total_num_gt + num_gt
        total_num_det = total_num_det + num_det

        local_accumulative_recall = 0
        local_accumulative_precision = 0
        gt_flag = np.zeros((1, num_gt))
        det_flag = np.zeros((1, num_det))

        #######first check for one-to-one case##########
        local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, \
        gt_flag, det_flag, hit_str_num = one_to_one(local_sigma_table, local_tau_table,
                                                    local_accumulative_recall, local_accumulative_precision,
                                                    global_accumulative_recall, global_accumulative_precision,
                                                    gt_flag, det_flag, idx)

        hit_str_count += hit_str_num
        #######then check for one-to-many case##########
        local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, \
        gt_flag, det_flag, hit_str_num = one_to_many(local_sigma_table, local_tau_table,
                                                     local_accumulative_recall, local_accumulative_precision,
                                                     global_accumulative_recall, global_accumulative_precision,
                                                     gt_flag, det_flag, idx)
        hit_str_count += hit_str_num
        #######then check for many-to-one case##########
        local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, \
        gt_flag, det_flag, hit_str_num = many_to_one(local_sigma_table, local_tau_table,
                                                     local_accumulative_recall, local_accumulative_precision,
                                                     global_accumulative_recall, global_accumulative_precision,
                                                     gt_flag, det_flag, idx)

        hit_str_count += hit_str_num

        # fid = open(fid_path, 'a+')
        try:
            local_precision = local_accumulative_precision / num_det
        except ZeroDivisionError:
            local_precision = 0

        try:
            local_recall = local_accumulative_recall / num_gt
        except ZeroDivisionError:
            local_recall = 0

        try:
            local_f_score = 2 * local_precision * local_recall / (
                local_precision + local_recall)
        except ZeroDivisionError:
            local_f_score = 0

    single_data['sigma'] = global_sigma
    single_data['global_tau'] = global_tau
    single_data['global_pred_str'] = global_pred_str
    single_data['global_gt_str'] = global_gt_str
    single_data["recall"] = local_recall
    single_data['precision'] = local_precision
    single_data['f_score'] = local_f_score
    return single_data


def combine_results(all_data):
    tr = 0.7
    tp = 0.6
    fsc_k = 0.8
    k = 2
    global_sigma = []
    global_tau = []
    global_pred_str = []
    global_gt_str = []
    for data in all_data:
        global_sigma.append(data['sigma'][0])
        global_tau.append(data['global_tau'][0])
        global_pred_str.append(data['global_pred_str'][0])
        global_gt_str.append(data['global_gt_str'][0])

    global_accumulative_recall = 0
    global_accumulative_precision = 0
    total_num_gt = 0
    total_num_det = 0
    hit_str_count = 0
    hit_count = 0

    def one_to_one(local_sigma_table, local_tau_table,
                   local_accumulative_recall, local_accumulative_precision,
                   global_accumulative_recall, global_accumulative_precision,
                   gt_flag, det_flag, idy):
        hit_str_num = 0
        for gt_id in range(num_gt):
            gt_matching_qualified_sigma_candidates = np.where(
                local_sigma_table[gt_id, :] > tr)
            gt_matching_num_qualified_sigma_candidates = gt_matching_qualified_sigma_candidates[
                0].shape[0]
            gt_matching_qualified_tau_candidates = np.where(
                local_tau_table[gt_id, :] > tp)
            gt_matching_num_qualified_tau_candidates = gt_matching_qualified_tau_candidates[
                0].shape[0]

            det_matching_qualified_sigma_candidates = np.where(
                local_sigma_table[:, gt_matching_qualified_sigma_candidates[0]]
                > tr)
            det_matching_num_qualified_sigma_candidates = det_matching_qualified_sigma_candidates[
                0].shape[0]
            det_matching_qualified_tau_candidates = np.where(
                local_tau_table[:, gt_matching_qualified_tau_candidates[0]] >
                tp)
            det_matching_num_qualified_tau_candidates = det_matching_qualified_tau_candidates[
                0].shape[0]

            if (gt_matching_num_qualified_sigma_candidates == 1) and (gt_matching_num_qualified_tau_candidates == 1) and \
                    (det_matching_num_qualified_sigma_candidates == 1) and (
                    det_matching_num_qualified_tau_candidates == 1):
                global_accumulative_recall = global_accumulative_recall + 1.0
                global_accumulative_precision = global_accumulative_precision + 1.0
                local_accumulative_recall = local_accumulative_recall + 1.0
                local_accumulative_precision = local_accumulative_precision + 1.0

                gt_flag[0, gt_id] = 1
                matched_det_id = np.where(local_sigma_table[gt_id, :] > tr)
                # recg start
                gt_str_cur = global_gt_str[idy][gt_id]
                pred_str_cur = global_pred_str[idy][matched_det_id[0].tolist()[
                    0]]
                if pred_str_cur == gt_str_cur:
                    hit_str_num += 1
                else:
                    if pred_str_cur.lower() == gt_str_cur.lower():
                        hit_str_num += 1
                # recg end
                det_flag[0, matched_det_id] = 1
        return local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, gt_flag, det_flag, hit_str_num

    def one_to_many(local_sigma_table, local_tau_table,
                    local_accumulative_recall, local_accumulative_precision,
                    global_accumulative_recall, global_accumulative_precision,
                    gt_flag, det_flag, idy):
        hit_str_num = 0
        for gt_id in range(num_gt):
            # skip the following if the groundtruth was matched
            if gt_flag[0, gt_id] > 0:
                continue

            non_zero_in_sigma = np.where(local_sigma_table[gt_id, :] > 0)
            num_non_zero_in_sigma = non_zero_in_sigma[0].shape[0]

            if num_non_zero_in_sigma >= k:
                ####search for all detections that overlaps with this groundtruth
                qualified_tau_candidates = np.where((local_tau_table[
                    gt_id, :] >= tp) & (det_flag[0, :] == 0))
                num_qualified_tau_candidates = qualified_tau_candidates[
                    0].shape[0]

                if num_qualified_tau_candidates == 1:
                    if ((local_tau_table[gt_id, qualified_tau_candidates] >= tp)
                            and
                        (local_sigma_table[gt_id, qualified_tau_candidates] >=
                         tr)):
                        # became an one-to-one case
                        global_accumulative_recall = global_accumulative_recall + 1.0
                        global_accumulative_precision = global_accumulative_precision + 1.0
                        local_accumulative_recall = local_accumulative_recall + 1.0
                        local_accumulative_precision = local_accumulative_precision + 1.0

                        gt_flag[0, gt_id] = 1
                        det_flag[0, qualified_tau_candidates] = 1
                        # recg start
                        gt_str_cur = global_gt_str[idy][gt_id]
                        pred_str_cur = global_pred_str[idy][
                            qualified_tau_candidates[0].tolist()[0]]
                        if pred_str_cur == gt_str_cur:
                            hit_str_num += 1
                        else:
                            if pred_str_cur.lower() == gt_str_cur.lower():
                                hit_str_num += 1
                        # recg end
                elif (np.sum(local_sigma_table[gt_id, qualified_tau_candidates])
                      >= tr):
                    gt_flag[0, gt_id] = 1
                    det_flag[0, qualified_tau_candidates] = 1
                    # recg start
                    gt_str_cur = global_gt_str[idy][gt_id]
                    pred_str_cur = global_pred_str[idy][
                        qualified_tau_candidates[0].tolist()[0]]
                    if pred_str_cur == gt_str_cur:
                        hit_str_num += 1
                    else:
                        if pred_str_cur.lower() == gt_str_cur.lower():
                            hit_str_num += 1
                    # recg end

                    global_accumulative_recall = global_accumulative_recall + fsc_k
                    global_accumulative_precision = global_accumulative_precision + num_qualified_tau_candidates * fsc_k

                    local_accumulative_recall = local_accumulative_recall + fsc_k
                    local_accumulative_precision = local_accumulative_precision + num_qualified_tau_candidates * fsc_k

        return local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, gt_flag, det_flag, hit_str_num

    def many_to_one(local_sigma_table, local_tau_table,
                    local_accumulative_recall, local_accumulative_precision,
                    global_accumulative_recall, global_accumulative_precision,
                    gt_flag, det_flag, idy):
        hit_str_num = 0
        for det_id in range(num_det):
            # skip the following if the detection was matched
            if det_flag[0, det_id] > 0:
                continue

            non_zero_in_tau = np.where(local_tau_table[:, det_id] > 0)
            num_non_zero_in_tau = non_zero_in_tau[0].shape[0]

            if num_non_zero_in_tau >= k:
                ####search for all detections that overlaps with this groundtruth
                qualified_sigma_candidates = np.where((
                    local_sigma_table[:, det_id] >= tp) & (gt_flag[0, :] == 0))
                num_qualified_sigma_candidates = qualified_sigma_candidates[
                    0].shape[0]

                if num_qualified_sigma_candidates == 1:
                    if ((local_tau_table[qualified_sigma_candidates, det_id] >=
                         tp) and
                        (local_sigma_table[qualified_sigma_candidates, det_id]
                         >= tr)):
                        # became an one-to-one case
                        global_accumulative_recall = global_accumulative_recall + 1.0
                        global_accumulative_precision = global_accumulative_precision + 1.0
                        local_accumulative_recall = local_accumulative_recall + 1.0
                        local_accumulative_precision = local_accumulative_precision + 1.0

                        gt_flag[0, qualified_sigma_candidates] = 1
                        det_flag[0, det_id] = 1
                        # recg start
                        pred_str_cur = global_pred_str[idy][det_id]
                        gt_len = len(qualified_sigma_candidates[0])
                        for idx in range(gt_len):
                            ele_gt_id = qualified_sigma_candidates[0].tolist()[
                                idx]
                            if ele_gt_id not in global_gt_str[idy]:
                                continue
                            gt_str_cur = global_gt_str[idy][ele_gt_id]
                            if pred_str_cur == gt_str_cur:
                                hit_str_num += 1
                                break
                            else:
                                if pred_str_cur.lower() == gt_str_cur.lower():
                                    hit_str_num += 1
                                break
                        # recg end
                elif (np.sum(local_tau_table[qualified_sigma_candidates,
                                             det_id]) >= tp):
                    det_flag[0, det_id] = 1
                    gt_flag[0, qualified_sigma_candidates] = 1
                    # recg start
                    pred_str_cur = global_pred_str[idy][det_id]
                    gt_len = len(qualified_sigma_candidates[0])
                    for idx in range(gt_len):
                        ele_gt_id = qualified_sigma_candidates[0].tolist()[idx]
624
                        if ele_gt_id not in global_gt_str[idy]:
J
Jethong 已提交
625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728
                            continue
                        gt_str_cur = global_gt_str[idy][ele_gt_id]
                        if pred_str_cur == gt_str_cur:
                            hit_str_num += 1
                            break
                        else:
                            if pred_str_cur.lower() == gt_str_cur.lower():
                                hit_str_num += 1
                                break
                    # recg end

                    global_accumulative_recall = global_accumulative_recall + num_qualified_sigma_candidates * fsc_k
                    global_accumulative_precision = global_accumulative_precision + fsc_k

                    local_accumulative_recall = local_accumulative_recall + num_qualified_sigma_candidates * fsc_k
                    local_accumulative_precision = local_accumulative_precision + fsc_k
        return local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, gt_flag, det_flag, hit_str_num

    for idx in range(len(global_sigma)):
        local_sigma_table = np.array(global_sigma[idx])
        local_tau_table = global_tau[idx]

        num_gt = local_sigma_table.shape[0]
        num_det = local_sigma_table.shape[1]

        total_num_gt = total_num_gt + num_gt
        total_num_det = total_num_det + num_det

        local_accumulative_recall = 0
        local_accumulative_precision = 0
        gt_flag = np.zeros((1, num_gt))
        det_flag = np.zeros((1, num_det))

        #######first check for one-to-one case##########
        local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, \
        gt_flag, det_flag, hit_str_num = one_to_one(local_sigma_table, local_tau_table,
                                                    local_accumulative_recall, local_accumulative_precision,
                                                    global_accumulative_recall, global_accumulative_precision,
                                                    gt_flag, det_flag, idx)

        hit_str_count += hit_str_num
        #######then check for one-to-many case##########
        local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, \
        gt_flag, det_flag, hit_str_num = one_to_many(local_sigma_table, local_tau_table,
                                                     local_accumulative_recall, local_accumulative_precision,
                                                     global_accumulative_recall, global_accumulative_precision,
                                                     gt_flag, det_flag, idx)
        hit_str_count += hit_str_num
        #######then check for many-to-one case##########
        local_accumulative_recall, local_accumulative_precision, global_accumulative_recall, global_accumulative_precision, \
        gt_flag, det_flag, hit_str_num = many_to_one(local_sigma_table, local_tau_table,
                                                     local_accumulative_recall, local_accumulative_precision,
                                                     global_accumulative_recall, global_accumulative_precision,
                                                     gt_flag, det_flag, idx)
    try:
        recall = global_accumulative_recall / total_num_gt
    except ZeroDivisionError:
        recall = 0

    try:
        precision = global_accumulative_precision / total_num_det
    except ZeroDivisionError:
        precision = 0

    try:
        f_score = 2 * precision * recall / (precision + recall)
    except ZeroDivisionError:
        f_score = 0

    try:
        seqerr = 1 - float(hit_str_count) / global_accumulative_recall
    except ZeroDivisionError:
        seqerr = 1

    try:
        recall_e2e = float(hit_str_count) / total_num_gt
    except ZeroDivisionError:
        recall_e2e = 0

    try:
        precision_e2e = float(hit_str_count) / total_num_det
    except ZeroDivisionError:
        precision_e2e = 0

    try:
        f_score_e2e = 2 * precision_e2e * recall_e2e / (
            precision_e2e + recall_e2e)
    except ZeroDivisionError:
        f_score_e2e = 0

    final = {
        'total_num_gt': total_num_gt,
        'total_num_det': total_num_det,
        'global_accumulative_recall': global_accumulative_recall,
        'hit_str_count': hit_str_count,
        'recall': recall,
        'precision': precision,
        'f_score': f_score,
        'seqerr': seqerr,
        'recall_e2e': recall_e2e,
        'precision_e2e': precision_e2e,
        'f_score_e2e': f_score_e2e
    }
    return final