-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathadaptive-threshold-main_v2.py
1100 lines (941 loc) · 53.8 KB
/
adaptive-threshold-main_v2.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
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# import FaceRecognition as FR
import cv2
import glob
import numpy as np
import pickle
import os, sys
import time
import random
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import scipy.stats
from sympy import S, symbols, printing
from facenet import facenet, detect_face
from scipy.optimize import minimize, minimize_scalar
from faceDetection_frozenGraph import TensoflowFaceDector
from myFACENET import FACENET_EMBEDDINGS
from tqdm import tqdm
from sklearn.linear_model import LinearRegression
from scipy.interpolate import make_interp_spline, BSpline, CubicSpline
import scipy.integrate as integrate
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
import adaptive_config as myconfig
font = {'size': 10}
plt.rc('font', **font)
linearReg = LinearRegression()
sys.setrecursionlimit(10**7)
print('Recusion Limit: {}'.format(sys.getrecursionlimit()))
OUTPUT_CSV_PATH = 'results_csv_{}_{}'.format(myconfig.GSIZE, myconfig.DATA_VERSION)
GALLERY_FVECTORS = myconfig.GSIZE
if not os.path.exists(OUTPUT_CSV_PATH):
os.makedirs(OUTPUT_CSV_PATH)
tDetector = TensoflowFaceDector(myconfig.PATH_TO_CKPT_FACE)
myfacenet = FACENET_EMBEDDINGS(myconfig.PATH_TO_CKPT_FACENET_512D_9967)
def serialize_dict(dict_, filename):
mydict = dict()
# Load the embeddings of known people or check if there are any new people need to add in the list
for image in glob.glob('known_persons/*'):
name = image.split('/')[-1].split('.')[0]
image = FR.load_image_file(image)
encoding = FR.face_encoding(image)[0]
mydict[name] = encoding
# Serialize dictionary in binary format
with open(filename, 'wb') as handle:
pickle.dump(dict_, handle, protocol=pickle.HIGHEST_PROTOCOL)
def deserialize_dict(filename):
# Deserialize dictionary
with open(filename, 'rb') as handle:
return pickle.load(handle)
def face_distance(face_encoding, face_to_compare):
"""
Given a list of face encodings, compare them to a known face encoding and get a euclidean distance
for each comparison face. The distance tells you how similar the faces are.
:param faces: List of face encodings to compare
:param face_to_compare: A face encoding to compare against
:return: A numpy ndarray with the distance for each face in the same order as the 'faces' array
"""
if len(face_encoding) == 0:
return np.empty((0))
return np.linalg.norm(face_encoding - face_to_compare, ord=None, axis=1)
def compare_faces(known_face_encodings, face_encoding_to_check, tolerance=0.9):
"""
Compare a list of face encodings against a candidate encoding to see if they match.
:param known_face_encodings: A list of known face encodings
:param face_encoding_to_check: A single face encoding to compare against the list
:param tolerance: How much distance between faces to consider it a match. Lower is more strict. 0.6 is typical best performance.
:return: A list of True/False values indicating which known_face_encodings match the face encoding to check
"""
return list(face_distance(known_face_encodings, face_encoding_to_check) <= tolerance)
def cosine_similarity(dict_encoding, current_encoding):
return (np.dot(dict_encoding, current_encoding)/(np.linalg.norm(dict_encoding)*np.linalg.norm(current_encoding)))
def cosine_distance(dict_encoding, current_encoding):
return (1 - np.dot(dict_encoding, current_encoding)/(np.linalg.norm(dict_encoding)*np.linalg.norm(current_encoding)))
def euclidean_distance(dict_encoding, current_encoding):
return (np.sqrt(np.sum((dict_encoding - current_encoding) ** 2)))
def facenet_encoding():
pass
def dlib_encoding():
pass
def get_facial_embeddings(cropped_faces):
return myfacenet.run(cropped_faces)
def detect_crop_face(frame, frame_width, frame_height):
boxes, scores, classes, num_detections = tDetector.run(frame)
boxes = np.squeeze(boxes)
scores = np.squeeze(scores)
face_locations = list()
faces_cropped = list()
face_boxes = list()
current_encoding = None
face_count = 0
valid_scores = [score for score in scores if score > 0.4]
num_faces = len(valid_scores)
# print('No. of faces: {}, score: {}'.format(num_faces, valid_scores))
for score, box in zip(scores, boxes):
if score > 0.7 and num_faces == 1:
# ymin, xmin, ymax, xmax = box
left = int(box[1]*frame_width)
top = int(box[0]*frame_height)
right = int(box[3]*frame_width)
bottom = int(box[2]*frame_height)
face_locations.append((top, right, bottom, left))
face_boxes.append([left, top, right, bottom])
cropped = frame[top:bottom, left:right]
# cv2.imwrite('test.jpg', cropped)
cropped = cv2.resize(cropped, (160,160), interpolation=cv2.INTER_LINEAR)
faces_cropped.append(facenet.prewhiten(cropped))
num_curr_faces = len(faces_cropped)
# eucliDist_matrix = np.zeros((num_curr_faces, num_of_identities))
if num_curr_faces > 0:
if myconfig.FACE_RECOGNITION == 'DLIB':
face_encoding = FR.face_encoding(frame, face_locations)
if myconfig.FACE_RECOGNITION == 'FACENET':
#***********************************************************************************************
current_encoding = myfacenet.run(faces_cropped)
### DLIB
# # Loop through each face in this frame of video
# for (top, right, bottom, left), face_encoding in zip(face_locations, face_encoding):
# # See if the face is a match for the known face(s)
# matches = FR.compare_faces(known_face_encodings, face_encoding)
# name = "Unknown"
# # If a match was found in known_face_encodings, just use the first one.
# # if True in matches:
# # first_match_index = matches.index(True)
# # name = known_face_names[first_match_index]
# # Or instead, use the known face with the smallest distance to the new face
# face_distances = FR.face_distance(known_face_encodings, face_encoding)
# best_match_index = np.argmin(face_distances)
# if matches[best_match_index]:
# name = known_face_names[best_match_index]
# Draw a label with a name below the face
# cv2.putText(frame, name, (left+5, top-15), cv2.FONT_HERSHEY_DUPLEX, 1.0, (255, 255, 255), 1)
#************************************************************************************************
# Draw a box around the face
# cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255),int(round(frame_height/150)), 2)
return frame, faces_cropped, face_boxes, current_encoding
def find_threshold_stats(mydirs):
mydict_list = list()
# print(mydirs)
for dir_name in mydirs:
images_count = len(list(f for f in glob.glob(os.path.join(myconfig.SRC_PATH, dir_name)+'/*')))
if images_count >= myconfig.IMAGE_NUM_MIN and images_count <= myconfig.IMAGE_NUM_MAX:
cropped_faces = list()
faces_embeddings = list()
mydict = dict()
count = 0
for file in glob.glob(os.path.join(myconfig.SRC_PATH, dir_name)+'/*'):
filename = file.split('/')[-1]
frame = cv2.imread(file)
frame_height, frame_width, _ = frame.shape
#check if there are two faces in the frame
frame, cropped, face_boxes, current_encoding = detect_crop_face(frame, frame_width, frame_height)
if cropped is not None and current_encoding is not None:
cropped_faces.append(cropped)
faces_embeddings.append(current_encoding)
count += 1
if len(faces_embeddings) > 0:
# faces_embeddings.append(get_facial_embeddings(cropped_faces))
mydict['face_id'] = dir_name
mydict['face_embeddings'] = faces_embeddings
mydict_list.append(mydict)
# Check if dictionary is empty
if len(mydict_list) == 0:
raise Exception("The size of the dictionary: mydict is 0")
else:
print('dictionary size: {}'.format(len(mydict_list)))
database_face_ids = list()
database_face_embeddings = list()
for dict_ in mydict_list:
# print(dict_['face_embeddings'].shape)
# # sys.exit(0)
database_face_ids.append(dict_['face_id'])
# database_face_embeddings.append(np.squeeze(dict_['face_embeddings'], axis=0))
database_face_embeddings.append(dict_['face_embeddings'])
# stats_menu = ['min', 'max', 'mean', 'std', 'variance']
threshold_auto_dict = dict()
threshold_cross_dict = dict()
threshold_dict = dict()
threshold_auto_min = list()
threshold_cross_min = list()
threshold_auto_avg = list()
threshold_cross_avg = list()
auto_distance = list()
cross_distance = list()
auto_similarity = list()
cross_similarity = list()
auto_min, cross_min = list(), list()
auto_max, cross_max = list(), list()
auto_avg, cross_avg = list(), list()
auto_var, cross_var = list(), list()
auto_std, cross_std = list(), list()
P_same = 0
P_diff = 0
for i, name, dict_encodings in tqdm(zip(np.arange(len(database_face_ids)),
database_face_ids,
database_face_embeddings)):
for j, dict_encodings_copy in zip(np.arange(len(database_face_ids)), database_face_embeddings):
for k, (dict_encoding) in enumerate(zip(dict_encodings)): #each image for that person
dict_encoding = np.squeeze(np.asarray(dict_encoding), axis=0)
for l, (dict_encoding_copy) in enumerate(zip(dict_encodings_copy)): #each images of gallery
dict_encoding_copy = np.squeeze(np.asarray(dict_encoding_copy), axis=0).transpose()
if i == j:
if k != l:
P_same += 1
auto_similarity.append(cosine_similarity(dict_encoding, dict_encoding_copy).tolist()[0][0])
else:
P_diff += 1
cross_similarity.append(cosine_similarity(dict_encoding, dict_encoding_copy).tolist()[0][0])
# auto_similarity = np.array(auto_similarity)
# cross_similarity = np.array(cross_similarity)
# print('Auto-similarity size: {}, cross-similarity size: {}'.format(auto_similarity.shape, cross_similarity.shape))
try:
auto_min.append(np.min(auto_similarity))
auto_max.append(np.max(auto_similarity))
auto_avg.append(np.mean(auto_similarity))
auto_var.append(np.var(auto_similarity))
auto_std.append(np.std(auto_similarity))
# auto_stats = [auto_min, auto_max, auto_avg, auto_var, auto_std]
# threshold_auto_min.append(auto_min)
# threshold_auto_avg.append(auto_avg)
cross_min.append(np.min(cross_similarity))
cross_max.append(np.max(cross_similarity))
cross_avg.append(np.mean(cross_similarity))
cross_var.append(np.var(cross_similarity))
cross_std.append(np.std(cross_similarity))
# cross_stats = [cross_min, cross_max, cross_avg, cross_var, cross_std]
# threshold_cross_min.append(cross_min)
# threshold_cross_avg.append(cross_avg)
# threshold_dict[name] = np.asarray([auto_stats, cross_stats], dtype=np.float32)
except ValueError: #raised if auto_dist and cross_dist are empty.
pass
# # calculate parameters
# sample_mean = mean(cross_similarity)
# sample_std = std(cross_similarity)
# print('Mean=%.3f, Standard Deviation=%.3f' % (sample_mean, sample_std))
# # define the distribution
# dist = norm(sample_mean, sample_std)
auto_mean = np.mean(auto_similarity)
auto_std = np.std(auto_similarity)
auto_var = np.var(auto_similarity)
auto_min = np.min(auto_similarity)
auto_max = np.max(auto_similarity)
cross_mean = np.mean(cross_similarity)
cross_std = np.std(cross_similarity)
cross_var = np.var(cross_similarity)
cross_min = np.min(cross_similarity)
cross_max = np.max(cross_similarity)
A = auto_var - cross_var
B = 2*(auto_mean*cross_var - cross_mean*auto_var)
C = np.square(cross_mean)*auto_var - np.square(auto_mean)*cross_var - 2*auto_var*cross_var*np.log(auto_std/cross_std)
coeff = [A, B, C]
roots = [x for x in np.roots(coeff) if x>-0 and x<=1]
# print(roots)
if PLOT_FLAG:
# Density Plot and Histogram of all arrival delays
fig1 = plt.figure(num='histogram-gaussian-curve')
ax1 = sns.distplot(auto_similarity, norm_hist=True, kde=True,
bins=int(180/5), color = 'blue',
hist_kws={'edgecolor':'black'},
label='auto_similarity',
kde_kws={'linewidth': 2, 'linestyle':'--', 'color':'blue', 'alpha':0.8})
ax1.set(xlabel='Threshold', ylabel='probability density')
# Density Plot and Histogram of all arrival delays
ax1 = sns.distplot(cross_similarity, norm_hist=True, kde=True,
bins=int(180/5), color = 'red',
hist_kws={'edgecolor':'black'},
label='cross_similarity',
kde_kws={'linewidth': 2, 'linestyle':'--', 'color':'red', 'alpha':0.8})
ax1.set(xlabel='Threshold', ylabel='density function')
auto_x = np.linspace(auto_min-0.1, auto_max+0.2, 1000)
auto_y = scipy.stats.norm.pdf(auto_x, auto_mean, auto_std)
# auto_ymin = np.min(auto_y)
# auto_ymax = np.max(auto_y)
# auto_y = (auto_y-auto_ymin)/(auto_ymax-auto_ymin)
ax1.plot(auto_x, auto_y, color='blue', linewidth=2, label='auto_gaussian')
cross_x = np.linspace(cross_min-0.1, cross_max+0.2, 1000)
cross_y = scipy.stats.norm.pdf(cross_x, cross_mean, cross_std)
# cross_ymin = np.min(cross_y)
# cross_ymax = np.max(cross_y)
# cross_y = (cross_y-cross_ymin)/(cross_ymax-cross_ymin)
ax1.plot(cross_x, cross_y, color='red', linewidth=2, label='cross_gaussian')
ax1.legend(loc='upper center', bbox_to_anchor=(0.5, -0.12), fancybox=True, shadow=False, ncol=4)
# h1, l1 = ax1.get_legend_handles_labels()
# h2, l2 = ax2.get_legend_handles_labels()
# ax.legend(h1+h2, l1+l2, loc='upper center', bbox_to_anchor=(0.5, -0.1), fancybox=True, shadow=False, ncol=6)
# # f = [np.exp(-np.square(x-mean)/2*variance)/(np.sqrt(2*np.pi*variance)) for x in auto_similarity]
# # plt.plot(np.arange(len(auto_similarity)), f)
# # auto_var = np.var(auto_similarity)
# # cross_var = np.var(cross_similarity)
# # k1 = auto_var/(auto_var+cross_var)
# # k2 = cross_var/(auto_var+cross_var)
# # updated_similarity = k2*np.array(auto_similarity) + k1*np.array(cross_similarity)
# # Density Plot and Histogram of all arrival delays
# sns.distplot(updated_similarity, hist=True, kde=True,
# bins=int(180/5), color = 'green',
# hist_kws={'edgecolor':'black'},
# kde_kws={'linewidth': 4})
# fig1.subplots_adjust(bottom=0.2)
fig1.tight_layout()
fig1.savefig(myconfig.PLOT_PATH+'/histogram-gaussian-curve.pdf', bbox_inches='tight', dpi=300)
# sys.exit(0)
fig2 = plt.figure(num='auto-similarity distribution')
plt.scatter(np.arange(len(auto_similarity)), auto_similarity, s=1, label='auto_similarity', color='deepskyblue')
plt.plot(np.arange(len(auto_similarity)), [auto_avg]*len(auto_similarity), label='auto_average', linewidth=1, color='green')
plt.legend(loc='upper center', bbox_to_anchor=(0.5, -0.12), fancybox=True, shadow=False, ncol=2)
plt.xlabel('No. of similarity pairing')
plt.ylabel('Cosine similarity')
# plt.legend(loc="best")
# auto_dist_1 = np.asarray(auto_dist_1)
fig2.tight_layout()
fig2.savefig(myconfig.PLOT_PATH+'/auto-similarity distribution.pdf', bbox_inches='tight', dpi=300)
fig3 = plt.figure(num='cross-similarity distribution')
plt.scatter(np.arange(len(cross_similarity)), cross_similarity, s=1, label='cross_similarity', color='coral')
plt.plot(np.arange(len(cross_similarity)), [cross_avg]*len(cross_similarity), label='cross_average', linewidth=1, color='green')
plt.legend(loc='upper center', bbox_to_anchor=(0.5, -0.12), fancybox=True, shadow=False, ncol=2)
plt.xlabel('No. of similarity pairing')
plt.ylabel('Cosine similarity')
# plt.legend(loc="best")
# cross_dist_1 = np.asarray(cross_dist_1)
fig3.tight_layout()
fig3.savefig(myconfig.PLOT_PATH+'/cross-similarity distribution.pdf', bbox_inches='tight', dpi=300)
# Creating histogram
# fig, axs = plt.subplots(1, 1, fimyconfig.GSIZE =(10, 7), tight_layout = True)
# axs.hist(auto_similarity, bins = 10)
# axs.hist(auto_similarity, bins = 10)
# fig32 = plt.figure('Histogram')
bins = np.linspace(-1, 2, 100)
# plt.figure('Histogram')
fig10, ax10_1 = plt.subplots(num='Histogram')
ax10_2 = ax10_1.twinx()
ax10_2.hist(auto_similarity, bins, alpha=0.7, label='auto_similarity', color='deepskyblue')
ax10_2.set_ylabel('Auto-similarity distribution', fontsize=10)
ax10_2.set_xlabel('Cosine similarity', fontsize=10)
# plt.figure('Cross similarity histogram')
ax10_1.hist(cross_similarity, bins, alpha=0.8, label='cross_similarity', color='coral')
ax10_1.set_ylabel('Cross-similarity distribution', fontsize=10)
ax10_1.set_xlabel('Cosine similarity', fontsize=10)
# plt.xlabel('Cosine similarity')
# plt.ylabel('No. of occurances')
# plt.legend(loc='upper center', bbox_to_anchor=(0.5, -0.12), fancybox=True, shadow=False, ncol=2)
h10_1, l10_1 = ax10_1.get_legend_handles_labels()
h10_2, l10_2 = ax10_2.get_legend_handles_labels()
ax10_2.legend(h10_1+h10_2, l10_1+l10_2, loc='upper center', bbox_to_anchor=(0.5, -0.12), fancybox=True, shadow=False, ncol=6)
fig10.savefig(myconfig.PLOT_PATH+'/Histogram.pdf', bbox_inches='tight', dpi=300)
# cross_dist_1 = np.asarray(cross_dist_1)
# plt.figure('auto cosine similarity')
# plt.plot(np.arange(len(auto_similarity)), [auto_avg]*len(auto_similarity), label='auto_average')
# # FINDING THE THRESHOLD
# myauto_min = np.min(auto_similarity)
# mycross_max = np.max(cross_similarity)
# myauto_avg = np.mean(auto_similarity)
# mycross_avg = np.mean(cross_similarity)
# if myauto_min >= mycross_max:
# adaptive_threshold = (myauto_min + mycross_max) / 2
# else:
# adaptive_threshold = myauto_min
# print('Threshold={}'.format(adaptive_threshold))
# print('Auto Similarity # = {}, Cross_similarity # = {}'.format(len(auto_similarity), len(cross_similarity)))
# plt.show()
# if np.max(threshold_auto_min) < np.min(threshold_cross_min): # safest selection
# threshold_face = max(threshold_auto_min)
# status = 'Best'
# else:
# if np.mean(threshold_auto_avg) < np.min(threshold_cross_avg):
# threshold_face = np.mean(threshold_auto_avg)
# status = 'Good'
# elif np.mean(threshold_auto_avg) < np.mean(threshold_cross_avg):
# threshold_face = np.mean(threshold_auto_avg)
# status = 'Fine'
# else:
# threshold_face = np.min(threshold_cross_avg) #risky selection among above criterions
# status = 'Risky'
# print('\n')
# print('Min of auto_avg: {}'.format(np.min(threshold_auto_avg)))
# print('Max of auto_avg: {}'.format(np.max(threshold_auto_avg))) # NO USE
# print('Min of cross_avg: {}'.format(np.min(threshold_cross_avg)))
# print('Max of cross_avg: {}'.format(np.max(threshold_cross_avg)))
# print('Avg of auto_avg: {}'.format(np.mean(threshold_auto_avg)))
# print('Avg of cross_avg: {}'.format(np.mean(threshold_cross_avg)))
# print('\n')
# print('Min of auto_min: {}'.format(np.min(threshold_auto_min))) # NO USE
# print('Max of auto_min: {}'.format(np.max(threshold_auto_min)))
# print('Avg of auto_min: {}'.format(np.mean(threshold_auto_min)))
# print('Min of cross_min: {}'.format(np.min(threshold_cross_min)))
# print('Max of cross_min: {}'.format(np.max(threshold_cross_min))) # NO USE
# print('Avg of cross_min: {}'.format(np.mean(threshold_cross_min)))
# print('\n')
# print('Threshold_face: {} ({})'.format(threshold_face, status))
# print('\n')
# file = open("face_threshold.txt","w")
# file.write('Min of auto_avg: {} \n'.format(np.min(threshold_auto_avg)))
# file.write('Max of auto_avg: {} \n'.format(np.max(threshold_auto_avg))) # NO USE
# file.write('Min of cross_avg: {} \n'.format(np.min(threshold_cross_avg)))
# file.write('Max of cross_avg: {} \n'.format(np.max(threshold_cross_avg)))
# file.write('Avg of auto_avg: {} \n'.format(np.mean(threshold_auto_avg)))
# file.write('Avg of cross_avg: {} \n'.format(np.mean(threshold_cross_avg)))
# file.write('\n')
# file.write('Min of auto_min: {} \n'.format(np.min(threshold_auto_min))) # NO USE
# file.write('Max of auto_min: {} \n'.format(np.max(threshold_auto_min)))
# file.write('Avg of auto_min: {} \n'.format(np.mean(threshold_auto_min)))
# file.write('Min of cross_min: {} \n'.format(np.min(threshold_cross_min)))
# file.write('Max of cross_min: {} \n'.format(np.max(threshold_cross_min))) # NO USE
# file.write('Avg of cross_min: {} \n'.format(np.mean(threshold_cross_min)))
# file.write('\n')
# file.write('Threshold_face: {} ({})'.format(threshold_face, status))
# file.close()
# # # name_list, threshold_stats = list(), list()
# # # 0->min, 1->max, 2->avg, 3->variance, 4->standard deviation
# # for name, details in threshold_dict.items():
# # # print(name)
# # threshold_stats = np.squeeze(details, axis=2).tolist()
# # auto_stats = threshold_stats[0]
# # cross_stats = threshold_stats[1]
# # # print(auto_stats)
# # # print(cross_stats)
return max(roots), auto_mean, cross_mean, P_same, P_diff
# def compare_embeddings():
# for i, current_encoding, facebox in zip(np.arange(len(current_encoding)).tolist(), current_encoding, face_boxes):
# dist_updated = list()
# for j, dict_name, dict_encodings in zip(np.arange(num_of_identities), known_face_names, known_face_encodings):
# dist = list()
# for dict_encoding in dict_encodings:
# # dist.append(euclidean_distance(dict_encoding, current_encoding))
# dist.append(cosine_similarity(dict_encoding, current_encoding))
# # eucliDist_matrix[i][j] = np.min(np.asarray(dist, dtype=np.float32))
# dist_updated.append(max(dist))
# if max(dist_updated) > 0.7:
# name = known_face_names[np.argmax(dist_updated)]
# else:
# name = 'Unknown'
def evaluate_model(query_dict, gallery_dict, adaptive_threshold, P_same, P_diff):
tp = 0
fp = 0
fn = 0
tn = 0
mygallery_list = list()
for q_key, q_values in tqdm(query_dict.items()):
# dist_updated = list()
for q_embedding in q_values:
known_names = list()
dist = list()
mygallery = dict()
for g_key, g_values in gallery_dict.items():
for g_embedding in g_values:
known_names.append(g_key)
# dist.append(euclidean_distance(dict_encoding, current_encoding))
dist.append(cosine_similarity(q_embedding, g_embedding))
# mygallery['q_key'] = q_key
# mygallery['known_names'] = known_names
# mygallery['dist'] = dist
# mygallery_list.append(mygallery)
# dist_updated.append(max(dist))
max_dist = max(dist)
name = known_names[np.argmax(dist)]
if max_dist >= adaptive_threshold:
if name == q_key:
tp += 1
else:
fp += 1
else:
if name == q_key:
fn += 1
else:
tn += 1
precision = tp/(tp+fp+1e-7)
recall = tp/(tp+fn+1e-7)
accuracy = (tp+tn)/(tp+tn+fp+fn)
f1score = 2*(precision*recall)/(precision+recall+1e-7)
# adaptive_precisions.append(precision)
# adaptive_recalls.append(recall)
# adaptive_f1scores.append(f1score)
# print('TP={}, FP={}, TN={}, FN={}'.format(tp, fp, tn, fn))
# VAL = tp/P_same
# FAR = fp/P_diff
tpr = tp/(tp+fn+1e-7)
fpr = fp/(fp+tn+1e-7)
return precision, recall, f1score, accuracy, tpr, fpr
def obj_func(x):
tp = 0
fp = 0
fn = 0
tn = 0
# for mydict_ in mygallery_list:
# known_names = mydict_['known_names']
# dist = mydict_['dist']
# q_key = mydict_['q_key']
for q_key, q_values in tqdm(query_dict.items()):
# dist_updated = list()
for q_embedding in q_values:
known_names = list()
dist = list()
mygallery = dict()
for g_key, g_values in gallery_dict.items():
for g_embedding in g_values:
known_names.append(g_key)
# dist.append(euclidean_distance(dict_encoding, current_encoding))
dist.append(cosine_similarity(q_embedding, g_embedding))
# dist_updated.append(max(dist))
max_dist = max(dist)
name = known_names[np.argmax(dist)]
if max_dist >= x:
if name == q_key:
tp += 1
else:
fp += 1
else:
if name == q_key:
fn += 1
else:
tn += 1
precision = tp/(tp+fp+1e-7)
recall = tp/(tp+fn+1e-7)
accuracy = (tp+tn)/(tp+tn+fp+fn)
f1score = 2*(precision*recall)/(precision+recall+1e-7)
# adaptive_precisions.append(precision)
# adaptive_recalls.append(recall)
# adaptive_f1scores.append(f1score)
tpr = abs(tp)/abs(tp+fn+1e-7) #VAL
fpr = abs(fp)/abs(fp+tn+1e-7) #FAR
# print(abs(tpr-fpr))
return (tpr-fpr)
# return f1score
def myquadeqn(mypoly, x):
return mypoly[0]*x**2+mypoly[1]*x+mypoly[2]
def main():
tstart = time.time()
global query_dict, gallery_dict, PLOT_FLAG
query_dict = dict()
gallery_dict = dict()
print('Creating gallery and query dictionary...')
auto_means = list()
cross_means = list()
updated_dirs = list()
directories = [mydir for mydir in os.listdir(myconfig.SRC_PATH)]
FOLDER_COUNT = 0
for dir_name in directories:
images_count = len(list(f for f in glob.glob(os.path.join(myconfig.SRC_PATH, dir_name)+'/*')))
if images_count >= myconfig.IMAGE_NUM_MIN and images_count <= myconfig.IMAGE_NUM_MAX:
FOLDER_COUNT += 1
updated_dirs.append(dir_name)
# print('folder count: ', FOLDER_COUNT)
# sys.exit(0)
adaptive_precisions, adaptive_recalls, adaptive_f1scores, adaptive_thresholds = list(), list(), list(), list()
fixed_thresholds_1, fixed_precisions_1, fixed_recalls_1, fixed_f1scores_1 = list(), list(), list(), list()
fixed_thresholds_2, fixed_precisions_2, fixed_recalls_2, fixed_f1scores_2 = list(), list(), list(), list()
fixed_thresholds_3, fixed_precisions_3, fixed_recalls_3, fixed_f1scores_3 = list(), list(), list(), list()
TPRs_adaptive, FPRs_adaptive = [0,0,0,0,0,0,1,1,1,1,1], [0,0,0,0,0,0,1,1,1,1,1]
TPRs_fixed_1, FPRs_fixed_1 = [0,0,0,0,0,0,1,1,1,1,1], [0,0,0,0,0,0,1,1,1,1,1]
TPRs_fixed_2, FPRs_fixed_2 = [0,0,0,0,0,0,1,1,1,1,1], [0,0,0,0,0,0,1,1,1,1,1]
TPRs_fixed_3, FPRs_fixed_3 = [0,0,0,0,0,0,1,1,1,1,1], [0,0,0,0,0,0,1,1,1,1,1]
# TPRs_adaptive, FPRs_adaptive = list(), list()
# TPRs_fixed_1, FPRs_fixed_1 = list(), list()
# TPRs_fixed_2, FPRs_fixed_2 = list(), list()
# TPRs_fixed_3, FPRs_fixed_3 = list(), list()
loop_count = 0
dir_count = 0
PLOT_FLAG = False
# precision = 0
# recall = 0
# f1score = 0
ptotal, pcount, pfixed_count_1, pfixed_count_2, pfixed_count_3 = 0, 0, 0, 0, 0
mydirs = list()
num_people = 0
# OPTIMIZATION_METRIC = f1score
fixed_threshold_min = 0.3
fixed_threshold_default = 0.5
fixed_threshold_max = 0.7
print('UPDARED DIRS SIZE: {}'.format(len(updated_dirs)))
for d in updated_dirs:
loop_count += 1
if loop_count == len(updated_dirs):
PLOT_FLAG = True
num_people += 1
if dir_count < 1:
mydirs.append(d)
dir_count += 1
# continue
else:
mydirs.append(d)
for mydir in mydirs:
query_embeddings = list()
for img in glob.glob(os.path.join(myconfig.SRC_PATH, mydir)+'/*'):
frame = cv2.imread(img)
try:
frame_height, frame_width, _ = frame.shape
frame, faces_cropped, face_boxes, current_encoding = detect_crop_face(frame,
frame_width,
frame_height)
if current_encoding is not None:
query_embeddings.append(current_encoding[0])
except AttributeError:
print("Image is None. Couldn't find shape!")
continue
#code to move to next frame
if len(query_embeddings) != 0:
# WE CAN INCREASE NUMBER OF IMAGES IN GALLERY TO INCREASE PRECISION AND ACCURACY
indexes = np.arange(len(query_embeddings)).tolist()
rand_indexes = random.sample(indexes, k=GALLERY_FVECTORS)
for item in rand_indexes:
indexes.remove(item)
query_dict[mydir] = [query_embeddings[i] for i in indexes]
gallery_dict[mydir] = [query_embeddings[i] for i in rand_indexes]
# FINDING OPTIMAL THRESHOLD VALUE
adaptive_threshold, auto_mean, cross_mean, P_same, P_diff = find_threshold_stats(mydirs)
# print(adaptive_threshold)
# adaptive_thresholds.append(adaptive_threshold)
auto_means.append(auto_mean)
cross_means.append(cross_mean)
# print('Adaptive threshold: {}'.format(adaptive_threshold))
# # SERIALIZING GALLERY AND QUERY EMBEDDINGS
# serialize_dict(gallery_dict, 'data/gallery_dict.pkl')
# serialize_dict(query_dict, 'data/query_dict.pkl')
# ADAPTIVE THRESHOLD
adaptive_precision, adaptive_recall, adaptive_f1score, adaptive_accuracy, adaptive_tpr, adaptive_fpr = evaluate_model(
query_dict, gallery_dict, adaptive_threshold, P_same, P_diff)
adaptive_threshold_prev = adaptive_threshold
adaptive_precision_prev = adaptive_precision
adaptive_recall_prev = adaptive_recall
adaptive_f1score_prev = adaptive_f1score
adaptive_tpr_prev = adaptive_tpr
adaptive_fpr_prev = adaptive_fpr
adaptive_accuracy_prev = adaptive_accuracy
ptotal += 1
if adaptive_f1score < myconfig.METRIC_BOUND:
# if False:
print('optimization going on ...')
myfunc = lambda x: -obj_func(x)
# res = minimize_scalar(myfunc, bounds=(cross_mean, auto_mean), method='bounded')
res = minimize_scalar(myfunc, bounds=(0, 1), method='bounded')
threshold_updated = res.x
adaptive_precision, adaptive_recall, adaptive_f1score, adaptive_accuracy, adaptive_tpr, adaptive_fpr = evaluate_model(
query_dict, gallery_dict, threshold_updated, P_same, P_diff)
if adaptive_f1score >= myconfig.METRIC_BOUND:
pcount += 1
if adaptive_f1score >= adaptive_f1score_prev:
print('Seletected optimized threshold!')
opt_threshold = threshold_updated
# print('cross mean={}, auto mean={}, opt_threshold={}'.format(cross_mean, auto_mean, opt_threshold))
else:
opt_threshold = adaptive_threshold_prev
adaptive_precision = adaptive_precision_prev
adaptive_recall = adaptive_recall_prev
adaptive_f1score = adaptive_f1score_prev
adaptive_tpr = adaptive_tpr_prev
adaptive_fpr = adaptive_fpr_prev
adaptive_accuracy = adaptive_accuracy_prev
else:
print('taking normal threshold ...')
pcount += 1
opt_threshold = adaptive_threshold
# Evaluating model after searching optimal threshold
# precision, recall, f1score, accuracy, tpr, fpr = evaluate_model(query_dict, gallery_dict, opt_threshold, P_same, P_diff)
adaptive_precisions.append(adaptive_precision)
adaptive_recalls.append(adaptive_recall)
adaptive_f1scores.append(adaptive_f1score)
adaptive_thresholds.append(opt_threshold)
if adaptive_fpr>0 and adaptive_fpr<1:
TPRs_adaptive.append(adaptive_tpr)
FPRs_adaptive.append(adaptive_fpr)
elif adaptive_fpr == 0:
TPRs_adaptive.append(0)
FPRs_adaptive.append(0)
elif adaptive_fpr == 1:
TPRs_adaptive.append(1)
FPRs_adaptive.append(1)
# FIXED THRESHOLD
# fixed threshold --> lower bound (1)
precision_fixed_1, recall_fixed_1, f1score_fixed_1,accuracy_fixed1, tpr_fixed_1, fpr_fixed_1 = evaluate_model(
query_dict, gallery_dict, fixed_threshold_min, P_same, P_diff)
fixed_precisions_1.append(precision_fixed_1)
fixed_recalls_1.append(recall_fixed_1)
fixed_f1scores_1.append(f1score_fixed_1)
fixed_thresholds_1.append(fixed_threshold_min)
if fpr_fixed_1>0 and fpr_fixed_1<1:
TPRs_fixed_1.append(tpr_fixed_1)
FPRs_fixed_1.append(fpr_fixed_1)
elif fpr_fixed_1 == 0:
TPRs_fixed_1.append(0)
FPRs_fixed_1.append(0)
elif fpr_fixed_1 == 1:
TPRs_fixed_1.append(1)
FPRs_fixed_1.append(1)
if f1score_fixed_1 >= myconfig.METRIC_BOUND:
pfixed_count_1 += 1
# fixed threshold --> middle bound (2)
precision_fixed_2, recall_fixed_2, f1score_fixed_2, accuracy_fixed2, tpr_fixed_2, fpr_fixed_2 = evaluate_model(
query_dict, gallery_dict, fixed_threshold_default, P_same, P_diff)
fixed_precisions_2.append(precision_fixed_2)
fixed_recalls_2.append(recall_fixed_2)
fixed_f1scores_2.append(f1score_fixed_2)
if fpr_fixed_2>0 and fpr_fixed_2<1:
TPRs_fixed_2.append(tpr_fixed_2)
FPRs_fixed_2.append(fpr_fixed_2)
if fpr_fixed_2 == 0:
TPRs_fixed_2.append(0)
FPRs_fixed_2.append(0)
if fpr_fixed_2 == 1:
TPRs_fixed_2.append(1)
FPRs_fixed_2.append(1)
fixed_thresholds_2.append(fixed_threshold_default)
if f1score_fixed_2 >= myconfig.METRIC_BOUND:
pfixed_count_2 += 1
# fixed threshold --> upper bound (3)
precision_fixed_3, recall_fixed_3, f1score_fixed_3,accuracy_fixed3, tpr_fixed_3, fpr_fixed_3 = evaluate_model(
query_dict, gallery_dict, fixed_threshold_max, P_same, P_diff)
fixed_precisions_3.append(precision_fixed_3)
fixed_recalls_3.append(recall_fixed_3)
fixed_f1scores_3.append(f1score_fixed_3)
if fpr_fixed_3>0 and fpr_fixed_3<1:
TPRs_fixed_3.append(tpr_fixed_3)
FPRs_fixed_3.append(fpr_fixed_3)
if fpr_fixed_3 == 0:
TPRs_fixed_3.append(0)
FPRs_fixed_3.append(0)
if fpr_fixed_3 == 1:
TPRs_fixed_3.append(1)
FPRs_fixed_3.append(1)
fixed_thresholds_3.append(fixed_threshold_max)
if f1score_fixed_3 >= myconfig.METRIC_BOUND:
pfixed_count_3 += 1
# print(len(VALs))
# print(len(FARs))
# print('Final Optimum Threshold={}'.format(opt_threshold))
# print('VAL={}, FAR={}'.format(VAL*100, FAR*100))
# print('Precision={}, Recall={}, F1score={}'.format(precision*100, recall*100, f1score*100))
# print('PRECISION SIZE: {}'.format(len(adaptive_precisions)))
# print('Adaptive threshold size: {}'.format(len(adaptive_thresholds)))
df_adaptive = pd.DataFrame(list(zip(adaptive_thresholds, adaptive_precisions, adaptive_recalls, adaptive_f1scores)),
columns=['Adaptive-Threshold', 'Precision', 'Recall', 'F1-score'])
df_fixed_1 = pd.DataFrame(list(zip(fixed_thresholds_1, fixed_precisions_1, fixed_recalls_1, fixed_f1scores_1)),
columns=['Fixed-Threshold=0.3', 'Precision@threshold=0.3', 'Recall@threshold=0.3', 'F1-score@threshold=0.3'])
df_fixed_2 = pd.DataFrame(list(zip(fixed_thresholds_2, fixed_precisions_2, fixed_recalls_2, fixed_f1scores_2)),
columns=['Fixed-Threshold=0.5', 'Precision@threshold=0.5', 'Recall@threshold=0.5', 'F1-score@threshold=0.5'])
df_fixed_3 = pd.DataFrame(list(zip(fixed_thresholds_3, fixed_precisions_3, fixed_recalls_3, fixed_f1scores_3)),
columns=['Fixed-Threshold=0.7', 'Precision@threshold=0.7', 'Recall@threshold=0.7', 'F1-score@threshold=0.7'])
df_roc_adaptive = pd.DataFrame(list(zip(FPRs_adaptive, TPRs_adaptive)), columns=['FPR', 'TPR'])
df_roc_fixed_1 = pd.DataFrame(list(zip(FPRs_fixed_1, TPRs_fixed_1)), columns=['FPR@threshold=0.3', 'TPR@threshold=0.3'])
df_roc_fixed_2 = pd.DataFrame(list(zip(FPRs_fixed_2, TPRs_fixed_3)), columns=['FPR@threshold=0.5', 'TPR@threshold=0.5'])
df_roc_fixed_3 = pd.DataFrame(list(zip(FPRs_fixed_3, TPRs_fixed_3)), columns=['FPR@threshold=0.7', 'TPR@threshold=0.7'])
df_adaptive.to_csv('data/results_csv/adaptive_threshold.csv')
df_fixed_1.to_csv('data/results_csv/fixed_threshold_1.csv')
df_fixed_2.to_csv('data/results_csv/fixed_threshold_2.csv')
df_fixed_3.to_csv('data/results_csv/fixed_threshold_3.csv')
df_roc_adaptive.to_csv('data/results_csv/roc_curve_adaptive.csv')
df_roc_fixed_1.to_csv('data/results_csv/roc_curve@threshold=0.3.csv')
df_roc_fixed_2.to_csv('data/results_csv/roc_curve@threshold=0.5.csv')
df_roc_fixed_3.to_csv('data/results_csv/roc_curve@threshold=0.7.csv')
print('Percentage of precision@adaptive_threshold >= {}: {}%'.format(myconfig.METRIC_BOUND, pcount/(ptotal+1e-7)*100))
print('Percentage of precision@fixed_threshold=0.3 >= {}: {}%'.format(myconfig.METRIC_BOUND, pfixed_count_1/(ptotal+1e-7)*100))
print('Percentage of precision@fixed_threshold=0.5 >= {}: {}%'.format(myconfig.METRIC_BOUND, pfixed_count_2/(ptotal+1e-7)*100))
print('Percentage of precision@fixed_threshold=0.7 >= {}: {}%'.format(myconfig.METRIC_BOUND, pfixed_count_3/(ptotal+1e-7)*100))
print('**************MODEL ACCURACY*********************')
print('Percentage of accuracy@adaptive_threshold: {}%'.format(adaptive_accuracy*100))
print('Percentage of accuracy@fixed_threshold=0.3: {}%'.format(accuracy_fixed1*100))
print('Percentage of accuracy@fixed_threshold=0.5: {}%'.format(accuracy_fixed2*100))
print('Percentage of accuracy@fixed_threshold=0.7: {}%'.format(accuracy_fixed3*100))
print('Number of pople taken into account: {}'.format(num_people))
people_count = np.arange(len(adaptive_thresholds))+2
# print('People count: {}'.format(people_count))
#### PLOT - ROC CURVE -------------------------------------------------------------------------------
fig4, ax4 = plt.subplots(num='roc_curve')
ax4.scatter(FPRs_adaptive, TPRs_adaptive, s=50, color='blue')
ax4.scatter(FPRs_fixed_1, TPRs_fixed_1, s=50, marker='s', color='darkorange')
ax4.scatter(FPRs_fixed_2, TPRs_fixed_2, s=50, marker='*', color='green')
ax4.scatter(FPRs_fixed_3, TPRs_fixed_3, s=50, marker='^', color='red')
ax4.set_ylim([0.0,1.0])
ax4.set_xlim([0,1])
# print(len('TPRS_adaptive size={}, FPRs_adaptive size={}'.format(TPRs_adaptive, FPRs_adaptive)))
mypoly = np.polyfit(FPRs_adaptive, TPRs_adaptive, 2)
mymodel = np.poly1d(mypoly)
new_FPRs = np.linspace(0, 1, 200)
TPRs_predicted = mymodel(new_FPRs)
# Estimating ROC-area under the curve
x = new_FPRs
auc_adaptive = integrate.quad(lambda x: myquadeqn(mypoly, x), 0.0, 1.0)[0]
# print('AUC: {}'.format(auc))
ax4.plot(new_FPRs, TPRs_predicted, label='adaptive_threshold', linewidth=2, color='blue')
# mymodel1 = np.poly1d(np.polyfit(FPRs_fixed_1, TPRs_fixed_1, 2))
# TPRs_predicted_fixed_1 = mymodel1(new_FPRs)
# ax4.plot(new_FPRs, TPRs_predicted_fixed_1, label='fixed_threshold=0.3', linewidth=3, color='darkorange')
# mymodel2 = np.poly1d(np.polyfit(FPRs_fixed_2, TPRs_fixed_2, 2))
# TPRs_predicted_fixed_2 = mymodel2(new_FPRs)
# ax4.plot(new_FPRs, TPRs_predicted_fixed_2, label='fixed_threshold=0.5', linewidth=3, color='green')
# mymodel3 = np.poly1d(np.polyfit(FPRs_fixed_3, TPRs_fixed_3, 2))
# TPRs_predicted_fixed_3 = mymodel3(new_FPRs)
# ax4.plot(new_FPRs, TPRs_predicted_fixed_3, label='fixed_threshold=0.7', linewidth=3, color='red')
mypoly1 = np.polyfit(FPRs_fixed_1, TPRs_fixed_1, 2)
mymodel1 = np.poly1d(mypoly1)
TPRs_predicted_fixed_1 = mymodel1(new_FPRs)
auc_fixed1 = integrate.quad(lambda x: myquadeqn(mypoly1, x), 0.0, 1.0)[0]
ax4.plot(new_FPRs, TPRs_predicted_fixed_1, label='fixed_threshold=0.3', linewidth=2, color='darkorange', linestyle='dashed')
mypoly2 = np.polyfit(FPRs_fixed_2, TPRs_fixed_2, 2)
mymodel2 = np.poly1d(mypoly2)
TPRs_predicted_fixed_2 = mymodel2(new_FPRs)
auc_fixed2 = integrate.quad(lambda x: myquadeqn(mypoly2, x), 0.0, 1.0)[0]
ax4.plot(new_FPRs, TPRs_predicted_fixed_2, label='fixed_threshold=0.5', linewidth=2, color='green', linestyle='dashed')
# print(FPRs_fixed_3)
# print(TPRs_fixed_3)
# sys.exit(0)
mypoly3 = np.polyfit(FPRs_fixed_3, TPRs_fixed_3, 2)
mymodel3 = np.poly1d(mypoly3)
TPRs_predicted_fixed_3 = mymodel3(new_FPRs)
auc_fixed3 = integrate.quad(lambda x: myquadeqn(mypoly3, x), 0.0, 1.0)[0]
ax4.plot(new_FPRs, TPRs_predicted_fixed_3, label='fixed_threshold=0.7', linewidth=2, color='red', linestyle='dashed')
ax4.plot([0,1], [0,1], linestyle='solid', linewidth=2, color='black')
# ax4.plot([0,0], [0,1], linestyle='dashed')
# ax4.plot([0,1], [0,0], linestyle='dashed')
# plt.legend(loc='best')
ax4.set_ylabel('True Positive rate (TPR)', fontsize=12)
ax4.set_xlabel('False Positive rate (FPR)', fontsize=12)
ax4.legend(loc='upper center', bbox_to_anchor=(0.5, -0.12), fancybox=True, shadow=False, ncol=4)
fig4.tight_layout()
fig4.savefig(myconfig.PLOT_PATH+'/roc_curve.pdf', bbox_inches='tight', dpi=300)
print('AUC_adaptive:{}'.format(auc_adaptive))
print('AUC_fixed1:{}'.format(auc_fixed1))
print('AUC_fixed2:{}'.format(auc_fixed2))
print('AUC_fixed3:{}'.format(auc_fixed3))
# fig4, ax4 = plt.subplots()
# ax4.scatter(adaptive_recalls, adaptive_precisions)
# ax4.set_ylim([0,1])
# ax4.set_xlim([0,1])
# mymodel2 = np.poly1d(np.polyfit(adaptive_recalls, adaptive_precisions, 2))
# new_recalls = np.linspace(0, 1, 200)
# precisions_predicted = mymodel2(new_recalls)
# ax4.plot(new_recalls, precisions_predicted)
# ax4.plot([0,1], [0.5,0.5], linestyle='dashed')
# fig1, ax1 = plt.subplots()
# ax1.scatter(people_count, adaptive_thresholds, s=50, label='adaptive_threshold', color='blue')
# ax1.plot(people_count, cross_means, label='cross-mean', linestyle='dashed', color='magenta')
# ax1.plot(people_count, auto_means, label='auto-mean', linestyle='dashed', color='darkblue')
# ax1.fill_between(people_count, cross_means, auto_means, color='deepskyblue', alpha=0.2)
# ax2 = ax1.twinx()
# ax2.plot(people_count, VALs, label='VAL', color='green')
# ax2.plot(people_count, FARs, label='FAR', color='red')
# plt.legend(loc='best')
# fig1.tight_layout()
# PLOT - ADAPTIVE METRICS COMPARISION -------------------------------------------------------------------
fig5, ax51 = plt.subplots(num='adaptive-metrics-comparision', figsize=(9, 5))
ax51.scatter(people_count, adaptive_thresholds, s=50, label='adaptive_threshold', color='blue')
ax51.plot(people_count, cross_means, label='cross-mean', linestyle='dotted', color='magenta', linewidth=1.5)
ax51.plot(people_count, auto_means, label='auto-mean', linestyle='dotted', color='indigo', linewidth=1.5)
ax51.fill_between(people_count, cross_means, auto_means, color='deepskyblue', alpha=0.1)
ax51.set_ylabel('Threshold', fontsize=10)
ax51.set_xlabel('No. of Identities', fontsize=10)
# ax51.legend(loc='upper center', bbox_to_anchor=(0.5, -0.1), fancybox=True, shadow=False, ncol=6)
ax52 = ax51.twinx()
ax52.plot(people_count, adaptive_precisions, label='precision', color='blueviolet', linewidth=1.5)
ax52.plot(people_count, adaptive_recalls, label='recall', color='green', linewidth=1.5)
ax52.plot(people_count, adaptive_f1scores, label='f1-score', color='red', linewidth=1.5)
ax52.set_ylabel('Performance Metrics', fontsize=10)