forked from clovaai/CLEval
-
Notifications
You must be signed in to change notification settings - Fork 0
/
script.py
718 lines (589 loc) · 31.6 KB
/
script.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
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
110
111
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
199
200
201
202
203
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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
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
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
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
624
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
from itertools import permutations
import rrc_evaluation_funcs
import math
import numpy as np
import Polygon as polygon3
from shapely.geometry import Point
from file_utils import load_zip_file, decode_utf8
from validation import validate_data
from arg_parser import PARAMS
import concurrent.futures
from tqdm import tqdm
def granularity_score(num_splitted):
"""get granularity penalty given number of how many splitted"""
return max(num_splitted-1, 0) * PARAMS.GRANULARITY_PENALTY_WEIGHT
def get_element_total_length(l):
return sum([len(x) for x in l])
def harmonic_mean(score1, score2):
"""get harmonic mean value"""
if score1+score2 == 0:
return 0
else:
return (2*score1*score2) / (score1+score2)
def lcs(s1, s2):
"""Longeset Common Sequence between s1 & s2"""
# source: https://stackoverflow.com/questions/48651891/longest-common-subsequence-in-python
if len(s1) == 0 or len(s2) == 0:
return 0, ''
matrix = [["" for x in range(len(s2))] for x in range(len(s1))]
for i in range(len(s1)):
for j in range(len(s2)):
if s1[i] == s2[j]:
if i == 0 or j == 0:
matrix[i][j] = s1[i]
else:
matrix[i][j] = matrix[i-1][j-1] + s1[i]
else:
matrix[i][j] = max(matrix[i-1][j], matrix[i][j-1], key=len)
cs = matrix[-1][-1]
return len(cs), cs
class GlobalResult(object):
"""Object that holds each record of all samples."""
def __init__(self, with_e2e=False):
self.with_e2e = with_e2e # flags for calculate end-to-end or not
# recording stats for detection evaluation
self.det_correct_num_recall = 0 # CorrectNum for Recall
self.det_correct_num_precision = 0 # CorrectNum for Precision
self.chars_gt = 0 # TotalNum for Recall (Both detection ,end-to-end)
self.chars_det = 0 # TotalNum for Precisiion
# recording stats for end-to-end evaluation
self.e2e_correct_num_recall = 0 # CorrectNum for Recall
self.e2e_correct_num_precision = 0 # CorrectNum for Precision
self.chars_recog = 0 # TotalNum for Precision
# metadata information
self.num_splitted = 0
self.num_merged = 0
self.num_false_positive = 0
self.char_missed = 0
self.char_overlapped = 0
self.char_false_positive = 0
# # recording stats for end-to-end evaluation
self.e2e_correct_num_recall = 0 # CorrectNum for Recall
self.e2e_correct_num_precision = 0 # CorrectNum for Precision
self.chars_recog = 0 # TotalNum for Precision
self.e2e_char_missed = 0
self.e2e_char_false_positive = 0
self.e2e_recog_score_chars = 0
self.e2e_recog_score_correct_num = 0
def accumulate_stats(self, sample_dict):
"""accumulate single sample statistics in this object."""
self.det_correct_num_recall += sample_dict['det_correct_num_recall']
self.det_correct_num_precision += sample_dict['det_correct_num_precision']
self.chars_gt += sample_dict['chars_gt']
self.chars_det += sample_dict['chars_det']
self.num_splitted += sample_dict['num_splitted']
self.num_merged += sample_dict['num_merged']
self.num_false_positive += sample_dict['num_false_positive']
self.char_missed += sample_dict['char_missed']
self.char_overlapped += sample_dict['char_overlapped']
self.char_false_positive += sample_dict['char_false_positive']
if self.with_e2e:
self.e2e_correct_num_recall += sample_dict['e2e_correct_num_recall']
self.e2e_correct_num_precision += sample_dict['e2e_correct_num_precision']
self.chars_recog += sample_dict['chars_recog']
self.e2e_char_missed += sample_dict['e2e_char_missed']
self.e2e_char_false_positive += sample_dict['e2e_char_false_positive']
self.e2e_recog_score_chars += sample_dict['e2e_recog_score_chars']
self.e2e_recog_score_correct_num += sample_dict['e2e_recog_score_correct_num']
def to_dict(self):
"""make stats to dictionary."""
det_recall = 0 if self.chars_gt == 0 else self.det_correct_num_recall / self.chars_gt
det_precision = 0 if self.chars_det == 0 else self.det_correct_num_precision / self.chars_det
det_hmean = harmonic_mean(det_recall, det_precision)
result_dict = {
'Detection': {
'recall': det_recall,
'precision': det_precision,
'hmean': det_hmean
},
'Detection_Metadata': {
'num_merge': self.num_merged,
'num_split': self.num_splitted,
'num_false_pos': self.num_false_positive,
'char_miss': self.char_missed,
'char_overlap': self.char_overlapped,
'char_false_pos': self.char_false_positive
}
}
e2e_recall = 0 if self.chars_gt == 0 else self.e2e_correct_num_recall / self.chars_gt
e2e_precision = 0 if self.chars_recog == 0 else self.e2e_correct_num_precision / self.chars_recog
e2e_hmean = harmonic_mean(e2e_recall, e2e_precision)
e2e_recog_score = 0 if self.e2e_recog_score_chars == 0 else self.e2e_recog_score_correct_num / self.e2e_recog_score_chars
result_dict.update({
'EndtoEnd': {
'recall': e2e_recall,
'precision': e2e_precision,
'hmean': e2e_hmean,
'recognition_score': e2e_recog_score,
},
'EndtoEnd_Metadata': {
'num_merge': self.num_merged,
'num_split': self.num_splitted,
'num_false_pos': self.num_false_positive,
'char_miss': self.e2e_char_missed,
'char_false_pos': self.e2e_char_false_positive
}
})
return result_dict
class SampleResult(object):
"""Object that holds result of single sample."""
def __init__(self, with_e2e=False, with_recog_score=False):
self.with_e2e = with_e2e
self.with_recog_score = with_recog_score
self.det_recall = 0 # Detection Recall score
self.det_precision = 0 # Detection Precision score
self.det_hmean = 0 # Detection H-Mean score
self.chars_gt = 0 # TotalNum for Recall (Both detection ,end-to-end)
self.chars_det = 0 # TotalNum for Precisiion
self.det_correct_num_recall = 0 # CorrectNum for recall
self.det_correct_num_precision = 0 # CorrectNum for precision
# detection metadata information
self.num_splitted = 0
self.num_merged = 0
self.num_false_pos = 0
self.char_missed = 0
self.char_overlapped = 0
self.char_false_pos = 0
if self.with_e2e:
# create variables for end-to-end evaluation
self.e2e_recall = 0 # End-to-End Recall score
self.e2e_precision = 0 # End-to-End Precision score
self.e2e_hmean = 0 # End-to-End H-Mean score
self.e2e_recog_score = 0 # Recognition Score
self.chars_recog = 0
self.e2e_correct_num_precision = 0
self.e2e_correct_num_recall = 0
self.e2e_result_matrix = np.zeros([1,1])
self.e2e_char_missed = 0
self.e2e_char_false_pos = 0
self.e2e_recog_score_chars = 0
self.e2e_recog_score_correct_num = 0
# Array of GT keys marked as don't Care
self.gt_dont_care_indices = []
# Array of Detected keys matched with a don't Care GT
self.det_dont_care_indices = []
# matrix for storing matching information (M_ij)
self.match_matrix = np.zeros([1,1])
# pseudo character centers (gt_pcc_points[i][k] == (c_i)^k )
self.gt_pcc_points = []
self.gt_char_counts = []
# for counting how may PCC points included
self.pcc_count_matrix = [] # ( (m_ij)^k )
self.gt_pcc_checked = []
self.gt_pcc_count = []
# for storing area precision / sample scoring
self.area_precision_matrix = np.zeros([1,1])
self.det_result_matrix = np.zeros([1,1])
# for storing pairs to visualize on Web
# (just to give different color for case OO, OM, MO, not used for evaluation)
self.pairs = []
# Below variables are for web visualization
# evaluation logs
self.eval_log = ""
# character count matrix
self.character_counts = np.zeros(([1,1]))
def prepare_gt(self, gt_boxes):
"""prepare ground-truth boxes in evaluation format."""
self.gt_boxes = gt_boxes
for gt_idx, gt_box in enumerate(self.gt_boxes):
if not PARAMS.CASE_SENSITIVE:
gt_box.transcription = gt_box.transcription.upper()
if gt_box.is_dc:
self.gt_dont_care_indices.append(gt_idx)
self.gt_pcc_points.append(gt_box.pseudo_character_center())
self.eval_log += "GT polygons: " + str(len(self.gt_boxes)) + (" (" + str(len(self.gt_dont_care_indices)) + " don't care) \n")
# subtract overlapping gt area from don't care boxes
# Area(Don't care) - Area(Ground Truth):
for dc in self.gt_dont_care_indices:
for idx in list(set(range(len(self.gt_boxes))) - set(self.gt_dont_care_indices)):
if self.gt_boxes[idx] & self.gt_boxes[dc] > 0:
# TODO: currently, PCC exclusion for area overlapped with don't care is not considered.
self.gt_boxes[dc].subtract(self.gt_boxes[idx])
def prepare_det(self, det_boxes):
"""prepare detection results in evaluation format."""
self.det_boxes = det_boxes
for det_idx, det_box in enumerate(self.det_boxes):
if not PARAMS.CASE_SENSITIVE:
det_box.transcription = det_box.transcription.upper()
self.eval_log += "DET polygons: {}\n".format(str(len(self.det_boxes)))
def total_character_counts(self):
"""get TotalNum for detection evaluation."""
total_num_recall = 0
total_num_precision = 0
for gt_idx in range(len(self.gt_boxes)):
if gt_idx not in self.gt_dont_care_indices:
total_num_recall += len(self.gt_boxes[gt_idx].transcription)
for det_idx in range(len(self.det_boxes)):
if det_idx not in self.det_dont_care_indices:
total_num_precision += sum(self.pcc_count_matrix[gt_idx][det_idx])
return total_num_recall, total_num_precision
def get_false_positive_char_counts(self):
"""get FalsePositive for detection evaluation."""
fp_char_counts = 0
for det_idx in range(len(self.det_boxes)):
# no match with any GTs && not matched with don't care
if self.match_matrix.sum(axis=0)[det_idx] == 0 and det_idx not in self.det_dont_care_indices:
fp_char_counts += min(round(0.5+(1 / (1e-5 + self.det_boxes[det_idx].aspect_ratio()))), 10)
return fp_char_counts
def sort_detbox_order_by_pcc(self, gt_idx, det_indices):
"""sort detected box order by pcc information."""
char_len = len(self.gt_pcc_points[gt_idx])
not_ordered_yet = det_indices
ordered_indices = list()
for c in range(char_len):
if len(not_ordered_yet) == 1:
break
for det_id in not_ordered_yet:
if self.pcc_count_matrix[gt_idx][det_id][c] == 1:
ordered_indices.append(det_id)
not_ordered_yet.remove(det_id)
break
ordered_indices.append(not_ordered_yet[0])
return ordered_indices
def lcs_elimination(self, gt_idx, sorted_det_indices):
"""longest common sequence elimination by sorted detection boxes"""
standard_script = self.gtQuery[gt_idx]
lcs_length, lcs_string = lcs(standard_script, "".join(self.det_trans_not_found[idx] for idx in sorted_det_indices))
for c in lcs_string:
self.gt_trans_not_found[gt_idx] = self.gt_trans_not_found[gt_idx].replace(c, '', 1)
for det_idx in sorted_det_indices:
if not self.det_trans_not_found[det_idx].find(c) < 0:
self.det_trans_not_found[det_idx] = self.det_trans_not_found[det_idx].replace(c, '', 1)
break
return lcs_length
def calc_area_precision(self):
"""calculate area precision between each GTbox and DETbox"""
for gt_idx, gt_box in enumerate(self.gt_boxes):
det_char_counts = []
self.gt_pcc_checked.append(np.zeros(len(self.gt_pcc_points[gt_idx])))
for det_idx, det_box in enumerate(self.det_boxes):
intersected_area = gt_box & det_box
if det_box.area() > 0.0:
self.area_precision_matrix[gt_idx, det_idx] = intersected_area / det_box.area()
det_char_counts.append(np.zeros(len(self.gt_pcc_points[gt_idx])))
self.gt_char_counts.append(det_char_counts)
self.pcc_count_matrix.append(det_char_counts)
def calc_pcc_inclusion(self):
"""fill PCC counting matrix by iterating each GTbox and DETbox"""
for gt_id, gt_box in enumerate(self.gt_boxes):
pcc_points = gt_box.pseudo_character_center()
for det_id, det_box in enumerate(self.det_boxes):
for pcc_id, pcc_point in enumerate(pcc_points):
if det_box.is_inside(pcc_point[0], pcc_point[1]):
self.pcc_count_matrix[gt_id][det_id][pcc_id] = 1
def filter_det_dont_care(self):
"""Filter detection Don't care boxes"""
if len(self.gt_dont_care_indices) > 0:
for det_id in range(len(self.det_boxes)):
area_precision_sum = 0
for gt_id in self.gt_dont_care_indices:
if sum(self.pcc_count_matrix[gt_id][det_id]) > 0:
area_precision_sum += self.area_precision_matrix[gt_id][det_id]
if area_precision_sum > PARAMS.AREA_PRECISION_CONSTRAINT:
self.det_dont_care_indices.append(det_id)
else:
for gt_id in self.gt_dont_care_indices:
if self.area_precision_matrix[gt_id, det_id] > PARAMS.AREA_PRECISION_CONSTRAINT:
self.det_dont_care_indices.append(det_id)
break
self.eval_log += " (" + str(len(self.det_dont_care_indices)) + " don't care)\n" if len(self.det_dont_care_indices) > 0 else "\n"
def one_to_one_match(self, row, col):
"""One-to-One match condition"""
cont = 0
for j in range(len(self.area_precision_matrix[0])):
if sum(self.pcc_count_matrix[row][j]) > 0 and self.area_precision_matrix[row, j] >= PARAMS.AREA_PRECISION_CONSTRAINT:
cont = cont + 1
if cont != 1:
return False
cont = 0
for i in range(len(self.area_precision_matrix)):
if sum(self.pcc_count_matrix[i][col]) > 0 and self.area_precision_matrix[i, col] >= PARAMS.AREA_PRECISION_CONSTRAINT:
cont = cont + 1
if cont != 1:
return False
if sum(self.pcc_count_matrix[row][col]) > 0 and self.area_precision_matrix[row, col] >= PARAMS.AREA_PRECISION_CONSTRAINT:
return True
return False
def one_to_many_match(self, gt_id):
"""One-to-Many match condition"""
many_sum = 0
detRects = []
for det_idx in range(len(self.area_precision_matrix[0])):
if det_idx not in self.det_dont_care_indices:
if self.area_precision_matrix[gt_id, det_idx] >= PARAMS.AREA_PRECISION_CONSTRAINT and \
sum(self.pcc_count_matrix[gt_id][det_idx]) > 0:
many_sum += sum(self.pcc_count_matrix[gt_id][det_idx])
detRects.append(det_idx)
if many_sum > 0 and len(detRects) >= 2:
return True, detRects
else:
return False, []
def many_to_one_match(self, det_id):
"""Many-to-One match condition"""
many_sum = 0
gtRects = []
for gt_idx in range(len(self.area_precision_matrix)):
if gt_idx not in self.gt_dont_care_indices:
if sum(self.pcc_count_matrix[gt_idx][det_id]) > 0:
many_sum += self.area_precision_matrix[gt_idx][det_id]
gtRects.append(gt_idx)
if many_sum >= PARAMS.AREA_PRECISION_CONSTRAINT and len(gtRects) >= 2:
return True, gtRects
else:
return False, []
def calc_match_matrix(self):
"""Calculate match matrix with PCC counting matrix information."""
self.eval_log += "Find one-to-one matches\n"
for gt_id in range(len(self.gt_boxes)):
for det_id in range(len(self.det_boxes)):
if gt_id not in self.gt_dont_care_indices and det_id not in self.det_dont_care_indices:
match = self.one_to_one_match(gt_id, det_id)
if match:
self.pairs.append({'gt': [gt_id], 'det': [det_id], 'type': 'OO'})
self.eval_log += "Match GT #{} with Det #{}\n".format(gt_id, det_id)
# one-to-many match
self.eval_log += "Find one-to-many matches\n"
for gt_id in range(len(self.gt_boxes)):
if gt_id not in self.gt_dont_care_indices:
match, matched_det = self.one_to_many_match(gt_id)
if match:
self.pairs.append({'gt': [gt_id], 'det': matched_det, 'type': 'OM'})
self.eval_log += "Match GT #{} with Det #{}\n".format(gt_id, matched_det)
# many-to-one match
self.eval_log += "Find many-to-one matches\n"
for det_id in range(len(self.det_boxes)):
if det_id not in self.det_dont_care_indices:
match, matched_gt = self.many_to_one_match(det_id)
if match:
self.pairs.append({'gt': matched_gt, 'det': [det_id], 'type': 'MO'})
self.eval_log += "Match GT #{} with Det #{}\n".format(matched_gt, det_id)
for pair in self.pairs:
self.match_matrix[pair['gt'], pair['det']] = 1
# clear pcc count flag for not matched pairs
for gt_idx in range(len(self.gt_boxes)):
for det_idx in range(len(self.det_boxes)):
if not self.match_matrix[gt_idx][det_idx]:
for pcc in range(len(self.gt_pcc_points[gt_idx])):
self.pcc_count_matrix[gt_idx][det_idx][pcc] = 0
def eval_det(self):
self.eval_log += "<b>Detection | PRECISION\n</b>"
for detNum in range(len(self.det_boxes)):
box_precision = 0
if detNum in self.det_dont_care_indices:
continue
if self.match_matrix.sum(axis=0)[detNum] > 0:
matched_gt_indices = np.where(self.match_matrix[:, detNum] == 1)[0]
if len(matched_gt_indices) > 1:
self.num_merged += 1
for gt_idx in matched_gt_indices:
intermediate_precision = 0
found_char_pos = np.where(self.pcc_count_matrix[gt_idx][detNum] == 1)[0]
for x in found_char_pos:
if self.gt_pcc_checked[gt_idx][x] == 0:
self.gt_pcc_checked[gt_idx][x] = 1
box_precision += 1
intermediate_precision += 1
elif self.gt_pcc_checked[gt_idx][x] >= 1:
self.char_overlapped += 1
self.det_result_matrix[gt_idx][detNum] = intermediate_precision
self.det_result_matrix[len(self.gt_boxes)][detNum] = box_precision
self.det_result_matrix[len(self.gt_boxes)+1][detNum] = granularity_score(len(matched_gt_indices))
else:
self.num_false_pos += 1
# Recall score
self.eval_log += "<b>Detection | RECALL\n</b>"
for gtNum in range(len(self.gt_boxes)):
if gtNum in self.gt_dont_care_indices:
continue
found_gt_chars = 0
num_gt_characters = len(self.gt_pcc_points[gtNum])
box_char_recall_list = np.ones(num_gt_characters)
if self.match_matrix.sum(axis=1)[gtNum] > 0:
matched_det_indices = np.where(self.match_matrix[gtNum] > 0)[0]
if len(matched_det_indices) > 1:
self.num_splitted += 1
found_gt_chars = np.sum(self.gt_pcc_checked[gtNum])
self.char_missed += int(np.sum(box_char_recall_list) - found_gt_chars)
self.det_result_matrix[gtNum][len(self.det_boxes)+1] = granularity_score(len(matched_det_indices))
else:
self.char_missed += int(np.sum(box_char_recall_list))
self.det_result_matrix[gtNum][len(self.det_boxes)] = found_gt_chars
# Pseudo Character Center Visualization
self.character_counts = np.zeros((len(self.gt_boxes), len(self.det_boxes)))
for gtNum in range(len(self.gt_boxes)):
for detNum in range(len(self.det_boxes)):
self.character_counts[gtNum][detNum] = sum(self.gt_char_counts[gtNum][detNum])
# calculate precision / recall
self.chars_gt, self.chars_det = self.total_character_counts()
self.char_false_pos += self.get_false_positive_char_counts()
self.chars_det += self.char_false_pos
self.eval_log += "<b># of false positive chars\n</b>"
self.eval_log += "{}\n".format(self.char_false_pos)
# Sample Score : Character correct length - Granularity Penalty
self.det_correct_num_recall = max(np.sum(self.det_result_matrix[:, len(self.det_boxes)]) - np.sum(self.det_result_matrix[:, len(self.det_boxes)+1]), 0)
self.det_correct_num_precision = max(np.sum(self.det_result_matrix[len(self.gt_boxes)]) - np.sum(self.det_result_matrix[len(self.gt_boxes)+1]), 0)
self.det_recall = float(0) if self.chars_gt == 0 else float(self.det_correct_num_recall) / self.chars_gt
self.det_precision = float(0) if self.chars_det == 0 else float(self.det_correct_num_precision) / self.chars_det
self.det_hmean = harmonic_mean(self.det_recall, self.det_precision)
def eval_e2e(self):
self.gtQuery = [box.transcription for box in self.gt_boxes]
self.detQuery = [box.transcription for box in self.det_boxes]
self.gt_trans_not_found = [box.transcription for box in self.gt_boxes]
self.det_trans_not_found = [box.transcription for box in self.det_boxes]
self.eval_log += "=================================\n"
self.eval_log += "<b>End-to-End | Recall\n</b>"
for gtNum in range(len(self.gt_boxes)):
if gtNum in self.gt_dont_care_indices:
continue
if self.match_matrix.sum(axis=1)[gtNum] > 0:
matched_det_indices = np.where(self.match_matrix[gtNum] > 0)[0]
sorted_det_indices = self.sort_detbox_order_by_pcc(gtNum, matched_det_indices.tolist())
corrected_num_chars = self.lcs_elimination(gtNum, sorted_det_indices)
self.e2e_result_matrix[gtNum][len(self.det_boxes)] = corrected_num_chars
self.e2e_result_matrix[gtNum][len(self.det_boxes)+1] = granularity_score(len(matched_det_indices))
self.eval_log += "<b>End-to-End | Precision\n</b>"
for detNum in range(len(self.det_boxes)):
if detNum in self.det_dont_care_indices:
continue
if self.match_matrix.sum(axis=0)[detNum] > 0:
matched_gt_indices = np.where(self.match_matrix[:, detNum] == 1)[0]
self.e2e_result_matrix[len(self.gt_boxes)+1][detNum] = granularity_score(len(matched_gt_indices))
self.e2e_result_matrix[len(self.gt_boxes)][detNum] = len(self.detQuery[detNum]) - len(self.det_trans_not_found[detNum])
self.chars_recog = get_element_total_length([x for k, x in enumerate(self.detQuery) if k not in self.det_dont_care_indices])
# Sample Score : Character correct length - Granularity Penalty
self.e2e_correct_num_recall = max(np.sum(self.e2e_result_matrix[:, len(self.det_boxes)]) - np.sum(self.e2e_result_matrix[:, len(self.det_boxes)+1]), 0)
self.e2e_correct_num_precision = max(np.sum(self.e2e_result_matrix[len(self.gt_boxes)]) - np.sum(self.e2e_result_matrix[len(self.gt_boxes)+1]), 0)
self.e2e_char_missed = self.chars_gt - self.e2e_correct_num_recall
self.e2e_char_false_pos = self.chars_recog - np.sum(self.e2e_result_matrix[len(self.gt_boxes)])
self.e2e_recall = float(0) if self.chars_gt == 0 else float(self.e2e_correct_num_recall) / self.chars_gt
self.e2e_precision = float(0) if self.chars_recog == 0 else float(self.e2e_correct_num_precision) / self.chars_recog
self.e2e_hmean = harmonic_mean(self.e2e_recall, self.e2e_precision)
# Additional recognition score calculation. Notated as RS in paper.
for det in np.where(np.sum(self.match_matrix, axis=0) > 0)[0]:
self.e2e_recog_score_chars += len(self.detQuery[det])
self.e2e_recog_score_correct_num = np.sum(self.e2e_result_matrix[len(self.gt_boxes)])
self.e2e_recog_score = float(0) if self.e2e_recog_score_chars == 0 else float(self.e2e_recog_score_correct_num) / self.e2e_recog_score_chars
def evaluation(self):
self.area_precision_matrix = np.zeros([len(self.gt_boxes), len(self.det_boxes)])
self.det_result_matrix = np.zeros([len(self.gt_boxes)+2, len(self.det_boxes)+2])
self.match_matrix = np.zeros([len(self.gt_boxes), len(self.det_boxes)])
self.calc_area_precision()
self.calc_pcc_inclusion()
self.filter_det_dont_care()
self.calc_match_matrix()
# Matching Process
self.eval_det()
# Evaluation Process
if self.with_e2e:
self.e2e_result_matrix = np.zeros([len(self.gt_boxes)+2, len(self.det_boxes)+2])
self.eval_e2e()
def to_dict(self):
# print(self.pcc_count_matrix[0], type(self.pcc_count_matrix[0]))
# print(self.pcc_count_matrix[0][0], type(self.pcc_count_matrix[0][0]))
sample_metric_dict = {
'Rawdata': {
'det_correct_num_recall': self.det_correct_num_recall,
'det_correct_num_precision': self.det_correct_num_precision,
'chars_gt': self.chars_gt,
'chars_det': self.chars_det,
'num_splitted': self.num_splitted,
'num_merged': self.num_merged,
'num_false_positive': self.num_false_pos,
'char_missed': self.char_missed,
'char_overlapped': self.char_overlapped,
'char_false_positive': self.char_false_pos
},
'Detection': {
'precision': self.det_precision,
'recall': self.det_recall,
'hmean': self.det_hmean,
},
'pairs': self.pairs,
'detectionMat': [] if len(self.gt_boxes) > 100 else self.det_result_matrix.tolist(),
'precisionMat': [] if len(self.det_boxes) > 100 else self.area_precision_matrix.tolist(),
'gtPolPoints': [box.points for box in self.gt_boxes],
'detPolPoints': [box.points for box in self.det_boxes],
'gtCharPoints': self.gt_pcc_points,
'gtCharCounts': [np.sum(x, axis=0).tolist() for x in self.pcc_count_matrix],
'gtDontCare': self.gt_dont_care_indices,
'detDontCare': self.det_dont_care_indices,
'evaluationParams': vars(PARAMS),
'evaluationLog': self.eval_log
}
if self.with_e2e:
sample_metric_dict['Rawdata'].update({
'e2e_correct_num_recall': self.e2e_correct_num_recall,
'e2e_correct_num_precision': self.e2e_correct_num_precision,
'chars_recog': self.chars_recog,
'e2e_char_missed': self.e2e_char_missed,
'e2e_char_false_positive': self.e2e_char_false_pos,
'e2e_recog_score_chars': self.e2e_recog_score_chars,
'e2e_recog_score_correct_num': self.e2e_recog_score_correct_num
})
sample_metric_dict.update({
'EndtoEnd': {
'precision': self.e2e_precision,
'recall': self.e2e_recall,
'hmean': self.e2e_hmean,
'recognition_score': self.e2e_recog_score
},
'end2endMat': [] if len(self.gt_boxes) > 100 else self.e2e_result_matrix.tolist(),
'gtTrans': [box.transcription for box in self.gt_boxes],
'detTrans': [box.transcription for box in self.det_boxes],
'gtQuery': self.gtQuery,
'detQuery': self.detQuery
})
return sample_metric_dict
def eval_single_result(gt_file, det_file):
sample_result = SampleResult(PARAMS.E2E, PARAMS.RS)
# We suppose there always exist transcription information on end-to-end evaluation
if PARAMS.E2E:
PARAMS.TRANSCRIPTION = True
gt_boxes = rrc_evaluation_funcs.parse_single_file(gt_file, PARAMS.CRLF, PARAMS.BOX_TYPE, True, False)
sample_result.prepare_gt(gt_boxes)
det_boxes = rrc_evaluation_funcs.parse_single_file(det_file, PARAMS.CRLF, PARAMS.BOX_TYPE,
PARAMS.TRANSCRIPTION, PARAMS.CONFIDENCES)
sample_result.prepare_det(det_boxes)
sample_result.evaluation()
return sample_result.to_dict()
def cleval_evaluation(gt_file, submit_file):
"""
evaluate and returns the results
Returns with the following values:
- method (required) Global method metrics. Ex: { 'Precision':0.8,'Recall':0.9 }
- per_sample (optional) Per sample metrics. Ex: {'sample1' : { 'Precision':0.8,'Recall':0.9 },
'sample2' : { 'Precision':0.8,'Recall':0.9 }
"""
# storing overall result
overall_result = GlobalResult(PARAMS.E2E)
# to store per sample evaluation results
per_sample_metrics = {}
gt_files = load_zip_file(gt_file, PARAMS.GT_SAMPLE_NAME_2_ID)
submission_files = load_zip_file(submit_file, PARAMS.DET_SAMPLE_NAME_2_ID, True)
# prepare ThreadPool for multi-process
executor = concurrent.futures.ProcessPoolExecutor(max_workers=PARAMS.NUM_WORKERS)
futures = {}
bar_len = len(gt_files)
for file_idx in gt_files:
gt_file = rrc_evaluation_funcs.decode_utf8(gt_files[file_idx])
if file_idx in submission_files:
det_file = decode_utf8(submission_files[file_idx])
if det_file is None:
det_file = ""
else:
det_file = ""
future = executor.submit(eval_single_result, gt_file, det_file)
futures[future] = file_idx
with tqdm(total=bar_len) as pbar:
pbar.set_description("Integrating results...")
for future in concurrent.futures.as_completed(futures):
file_idx = futures[future]
result = future.result()
per_sample_metrics[file_idx] = result
overall_result.accumulate_stats(result['Rawdata'])
pbar.update(1)
executor.shutdown()
resDict = {'calculated': True, 'Message': '', 'method': overall_result.to_dict(), 'per_sample': per_sample_metrics}
return resDict
if __name__ == '__main__':
rrc_evaluation_funcs.main_evaluation(validate_data, cleval_evaluation)