-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathflow_source.py
1241 lines (896 loc) · 48.9 KB
/
flow_source.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
# 2023 Gabriel J. Diaz @ RIT
import json
import os
import sys
import numpy as np
import logging
import pickle
from tqdm import tqdm
import warnings
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas
import h5py
from tkinter import Tk
from tkinter.filedialog import askopenfilename
warnings.simplefilter(action='ignore', category=FutureWarning)
try:
os.add_dll_directory("D://opencvgpu//opencv_build_310//bin")
os.add_dll_directory("C://Program Files//NVIDIA GPU Computing Toolkit//CUDA//v11.8//bin")
import cv2
except:
import cv2
sys.path.append('core')
logger = logging.getLogger(__name__)
# These lines allow me to see logging.info messages in my jupyter cell output
logger.addHandler(logging.StreamHandler(stream=sys.stdout))
def show_image(image):
if type(image) == cv2.cuda_GpuMat:
image = image.download()
cv2.imshow('temp', image)
cv2.waitKey(0)
cv2.destroyAllWindows()
class video_source():
def __init__(self, file_path, out_parent_dir="flow_out"):
self.file_path = file_path
self.source_file_name = os.path.split(file_path)[-1].split('.')[0]
self.source_file_suffix = os.path.split(file_path)[-1].split('.')[1]
self.out_parent_dir = out_parent_dir
self.raw_frames_out_path = os.path.join(out_parent_dir, self.source_file_name, 'raw')
self.flow_frames_out_path = os.path.join(out_parent_dir, self.source_file_name, 'flow')
self.video_out_path = os.path.abspath(os.path.join(out_parent_dir, self.source_file_name))
self.magnitude_out_path = os.path.join(out_parent_dir, self.source_file_name, 'magnitude_data')
self.video_target_path = False
self.cumulative_mag_hist = None
self.magnitude_bins = None
self.hist_params = (100, 0, 40)
self.flow_algo = False
self.clahe = False
self.cuda_enabled = True
self.save_midpoint_images = False
self.streamline_points = None
self.current_gpu = False
self.previous_gpu = False
self.current_gray = False
self.previous_gray = False
def init_flow(self, algorithm, height_width=False):
# Even if self.cuda_is_enabled=True, some algos don't support cuda!
algo_supports_cuda = False
if algorithm == False:
return False
elif algorithm == "deepflow":
self.flow_algo = cv2.optflow.createOptFlow_DeepFlow()
elif algorithm == "farneback":
if not self.cuda_enabled:
self.flow_algo = cv2.optflow.createOptFlow_Farneback()
# self.flow_algo.setFastPyramids(False)
# self.flow_algo.setFlags(0)
self.flow_algo.setNumIters(10) #10
self.flow_algo.setNumLevels(25)
self.flow_algo.setPolyN(5) # 5
self.flow_algo.setPolySigma(1.1)
self.flow_algo.setPyrScale(10.5) # 0.5
self.flow_algo.setWinSize(3) # 13
else:
algo_supports_cuda = True
self.flow_algo = cv2.cuda_FarnebackOpticalFlow.create()
elif algorithm == "tvl1":
if not self.cuda_enabled:
logger.error('Non-cuda optical flow calculation not yet supported for ' + algorithm,
stack_info=True, exc_info=True)
sys.exit(1)
algo_supports_cuda = True
self.flow_algo = cv2.cuda_OpticalFlowDual_TVL1.create()
self.flow_algo.setNumScales(30) # (1/5)^N-1 def: 5
# self.flow_algo.setScaleStep(0.7) #
# self.flow_algo.setLambda(0.5) # default 0.15. smaller = smoother output.
# self.flow_algo.setScaleStep(0.7) # 0.8 by default. Not well documented. 0.7 did better with dots?
# self.flow_algo.setEpsilon(0.005) # def: 0.01
# self.flow_algo.setTau(0.5)
# self.flow_algo.setGamma(0.5) # def 0
elif algorithm == "brox":
if not self.cuda_enabled:
logger.error('Non-cuda optical flow calculation not yet supported for ' + algorithm,
stack_info=True, exc_info=True)
sys.exit(1)
algo_supports_cuda = True
self.flow_algo = cv2.cuda_BroxOpticalFlow.create()
# self.flow_algo.setFlowSmoothness() # def alpha 0.197
# self.flow_algo.setGradientConstancyImportance() # def gamma 0
# self.flow_algo.setInnerIterations() # def 5
# self.flow_algo.setOuterIterations() # def 150
self.flow_algo.setPyramidScaleFactor(5) # def 0
self.flow_algo.setSolverIterations(2) # def 0
elif algorithm == "pyrLK":
if not self.cuda_enabled:
logger.error('Non-cuda optical flow calculation not yet supported for ' + algorithm, stack_info=True,
exc_info=True)
sys.exit(1)
algo_supports_cuda = True
self.flow_algo = cv2.cuda_DensePyrLKOpticalFlow.create()
self.flow_algo.setMaxLevel(6) # default 3
self.flow_algo.setWinSize((41, 41)) # default 13, 13
#self.flow_algo.setNumIters(40) # 30 def
elif algorithm == "nvidia1" or algorithm == "nvidia":
algo_supports_cuda = True
params = {'perfPreset': cv2.cuda.NvidiaOpticalFlow_1_0_NV_OF_PERF_LEVEL_SLOW}
self.flow_algo = cv2.cuda.NvidiaOpticalFlow_1_0.create((height_width[1], height_width[0]), **params)
elif algorithm == "nvidia2":
algo_supports_cuda = True
params = {'perfPreset': cv2.cuda.NvidiaOpticalFlow_2_0_NV_OF_PERF_LEVEL_SLOW,
'outputGridSize': cv2.cuda.NvidiaOpticalFlow_2_0_NV_OF_OUTPUT_VECTOR_GRID_SIZE_1,
'hintGridSize': cv2.cuda.NvidiaOpticalFlow_2_0_NV_OF_HINT_VECTOR_GRID_SIZE_1,
'enableTemporalHints': True,
'enableCostBuffer': True}
self.flow_algo = cv2.cuda.NvidiaOpticalFlow_2_0_create((height_width[1], height_width[0]), **params)
else:
logger.error('Optical flow algorithm not implemented, or a typo', stack_info=True,exc_info=True)
sys.exit(1)
return algo_supports_cuda
def set_stream_dimensions(self, stream, visualize_as, height, width):
stream.width = width
if visualize_as == "hsv_stacked":
stream.height = height * 2
else:
stream.height = height
return stream
def get_hdf5_filepath(self, algorithm: str, gaze_centered=False):
if gaze_centered:
hdf5_out_name = self.source_file_name + '_' + algorithm + '_gaze-centered_flow.hdf5'
else:
hdf5_out_name = self.source_file_name + '_' + algorithm + '_flow.hdf5'
return os.path.join(self.video_out_path,hdf5_out_name)
def open_hdf5_file(self, algorithm: str, gaze_centered):
print( self.get_hdf5_filepath(algorithm, gaze_centered) )
if not os.path.exists(self.get_hdf5_filepath(algorithm, gaze_centered)):
logger.error('HDF5 flow file not found. Try calculating optic flow with that algorithm first, and double-check that the gaze-centered flag is set appropriately.', stack_info=True, exc_info=True)
sys.exit(1)
hdf5_flow = h5py.File(self.get_hdf5_filepath(algorithm, gaze_centered), 'r')
return hdf5_flow
def calculate_flow(self,
video_out_name=False,
algorithm="nvidia2",
gaze_centered=False,
preprocess_frames = True,
save_input_images=False,
save_output_images=False):
video_in = cv2.VideoCapture(self.file_path)
width = int(video_in.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(video_in.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = video_in.get(cv2.CAP_PROP_FPS)
num_frames = int(video_in.get(cv2.CAP_PROP_FRAME_COUNT))
width = int(video_in.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(video_in.get(cv2.CAP_PROP_FRAME_HEIGHT))
algo_supports_cuda = self.init_flow(algorithm, (height, width))
if algo_supports_cuda and self.cuda_enabled:
self.current_gpu = cv2.cuda_GpuMat()
self.previous_gpu = cv2.cuda_GpuMat()
# self.clahe = cv2.cuda.createCLAHE(clipLimit=1.0, tileGridSize=(10, 10))
count = 0
success = 1
hdf5_filepath = self.get_hdf5_filepath(algorithm,gaze_centered=gaze_centered)
with h5py.File(hdf5_filepath, 'w') as hdf5_file_out:
# hdf5_filepath['fps'] = fps
# hdf5_filepath['num_frames'] = num_frames
# hdf5_filepath['video_in'] = num_frames
flow_hdf5_dataset = hdf5_file_out.create_dataset("flow", (1, height, width, 2),
maxshape=(num_frames, height, width, 2), dtype='f',chunks=True,
compression="gzip", compression_opts=4, shuffle=True)
flow_hdf5_dataset.attrs['approx_fps'] = fps
flow_hdf5_dataset.attrs['num_source_frames'] = num_frames
h = int(video_in.get(cv2.CAP_PROP_FOURCC))
flow_hdf5_dataset.attrs['source_fourcc'] = h
flow_hdf5_dataset.attrs['source_codec'] = chr(h&0xff) + chr((h>>8)&0xff) + chr((h>>16)&0xff) + chr((h>>24)&0xff)
# ##############################
# # Iterate through frames
for index in tqdm(range(num_frames), desc="Saving flow to " + hdf5_filepath, unit='frames', total=num_frames):
success, frame = video_in.read()
if not success:
print(f'Frame {index}: video_in.read() unsuccessful')
continue
if index == 0:
prev_frame = frame
if preprocess_frames:
prev_frame = self.preprocess_frame(prev_frame)
# Apply histogram normalization.
if algo_supports_cuda and self.cuda_enabled:
self.previous_gpu.upload(prev_frame)
self.previous_gpu = cv2.cuda.cvtColor(self.previous_gpu, cv2.COLOR_BGR2GRAY)
# self.previous_gpu = self.clahe.apply(previous_gpu, cv2.cuda_Stream.Null())
else:
self.previous_gray = cv2.cvtColor(prev_frame, cv2.COLOR_BGR2GRAY)
# image2_gray = self.clahe.apply(image2_gray)
# Write out a blank first frame
image_out = np.zeros([np.shape(frame)[0], np.shape(frame)[1], 3], dtype=np.uint8)
# video_out.write(image_out)
continue
else:
# If is gaze data, recenter
if gaze_centered:
self.recenter_frame_on_gaze(frame, index)
current_bgr_processed = self.preprocess_frame(frame)
if algo_supports_cuda and self.cuda_enabled:
self.current_gpu.upload(current_bgr_processed)
self.current_gpu = cv2.cuda.cvtColor(self.current_gpu, cv2.COLOR_BGR2GRAY)
else:
self.current_gray = cv2.cvtColor(current_bgr_processed, cv2.COLOR_BGR2GRAY)
flow = self.calculate_flow_for_frame(index,algo_supports_cuda)
# Save optical flow to HDF5
flow_hdf5_dataset.resize((flow_hdf5_dataset.shape[0] + 1, flow.shape[0], flow.shape[1], 2))
flow_hdf5_dataset[index] = flow
if self.cuda_enabled:
self.previous_gpu = self.current_gpu.clone()
self.previous_gray = self.current_gray
# Save input images?
if save_input_images:
if os.path.isdir(self.raw_frames_out_path) is False:
os.makedirs(self.raw_frames_out_path)
cv2.imwrite(str(os.path.join(self.raw_frames_out_path, 'frame-{}.png'.format(index))),
current_bgr_processed)
# Save output images?
if save_output_images:
if os.path.isdir(self.flow_frames_out_path) is False:
os.makedirs(self.flow_frames_out_path)
cv2.imwrite(str(os.path.join(self.flow_frames_out_path, 'frame-{}.png'.format(index))),
image_out)
video_in.release()
def calculate_flow_for_frame(self, index, algo_supports_cuda):
if self.cuda_enabled and algo_supports_cuda:
if type(self.flow_algo) == cv2.cuda.BroxOpticalFlow:
self.current_gpu.upload(self.current_gpu.download().astype('float32'))
self.previous_gpu.upload(self.previous_gpu.download().astype('float32'))
flow = self.flow_algo.calc(self.current_gpu, self.previous_gpu, None)
else:
flow = self.flow_algo.calc(self.current_gpu, self.previous_gpu, None)
if type(self.flow_algo) == cv2.cuda.NvidiaOpticalFlow_1_0:
flow = self.flow_algo.upSampler(flow[0],(self.current_gpu.size()[0],self.current_gpu.size()[1]), self.flow_algo.getGridSize(),None)
if type(self.flow_algo) == cv2.cuda.NvidiaOpticalFlow_2_0:
flow = self.flow_algo.convertToFloat(flow[0], None)
flow = flow.download()
else:
flow = self.flow_algo.calc(self.current_gray, self.previous_gray, None)
flow = flow[0]
return flow
# def flow_to_mag_angle(self, index, flow, lower_mag_threshold=False, upper_mag_threshold=False):
#
# # Convert flow to mag / angle
# magnitude, angle = cv2.cartToPolar(flow[..., 0], flow[..., 1])
# magnitude = self.filter_magnitude(magnitude) # 29.6,
# self.append_to_mag_histogram(index, magnitude) # 7.3
#
# magnitude = self.filter_magnitude(magnitude) # 29.6,
# magnitude = self.apply_magnitude_thresholds_and_rescale(magnitude,
# lower_mag_threshold=lower_mag_threshold,
# upper_mag_threshold=upper_mag_threshold)
#
# return magnitude, angle
def save_out_mag_histogram(self, algorithm):
# Save out magnitude data pickle and image
if os.path.isdir(self.magnitude_out_path) is False:
os.makedirs(self.magnitude_out_path)
mag_pickle_filename = self.source_file_name + '_' + algorithm + '_mag.pickle'
dbfile = open(os.path.join(self.magnitude_out_path, mag_pickle_filename), 'wb')
pickle.dump({"values": self.cumulative_mag_hist, "bins": self.magnitude_bins[1]}, dbfile)
dbfile.close()
mag_image_filename = self.source_file_name + '_' + algorithm + '_mag.jpg'
mag_image_fileloc = os.path.join(self.magnitude_out_path, mag_image_filename)
# self.generate_mag_histogram(mag_image_fileloc, cumulative_mag_hist, mag_hist[1])
import matplotlib.pyplot as plt
# This prevents a warning that was present on some versions of mpl
from matplotlib import use as mpl_use
mpl_use('qtagg')
fig, ax = plt.subplots(figsize=(8, 4))
# ax.bar(bins[:-1], mag_values, width = .75 * (bins[1]-bins[0]))
ax.bar(self.magnitude_bins[:-1], np.cumsum(self.cumulative_mag_hist) / sum(self.cumulative_mag_hist))
# width=.75 * (self.magnitude_bins[-1] - self.magnitude_bins[0]))
# tidy up the figure
ax.grid(True)
# ax.set_title('flow magnitude')
ax.set_xlabel('vector length')
ax.set_ylabel('likelihood')
print( f'Magnitude saved to {self.magnitude_out_path}')
plt.savefig(mag_image_fileloc)
def calculate_magnitude_distribution(self,algorithm, gaze_centered=False, stride_length = 20):
# Calculates magnitude distribution of flow vectors for every stride_length frame in central 2/3 of video
if isinstance(self, pupil_labs_source) == False & gaze_centered == True:
logger.error('Can\'t perform gaze centered operations for video_sources. Ignoring request!',stack_info=True, exc_info=True)
sys.exit(1)
def append_to_mag_histogram(index, magnitude):
mag_hist = np.histogram(magnitude, self.hist_params[0], (self.hist_params[1], self.hist_params[2]))
# Store the histogram of avg magnitudes
if index == 0:
# Store the first flow histogram
self.cumulative_mag_hist = mag_hist[0]
self.magnitude_bins = mag_hist[1]
else:
# Calc cumulative avg flow magnitude by adding the first flow histogram in a weighted manner
cumulative_mag_hist = np.divide(
np.sum([np.multiply((index - 1), self.cumulative_mag_hist), mag_hist[0]], axis=0), index)
print('Calculating magnitude distribution.')
hdf5_object = self.open_hdf5_file(algorithm,gaze_centered)
flow = hdf5_object['flow']
num_rows = np.shape(flow)[0]
one_third = int(num_rows/3)
row_indices = np.arange(one_third,num_rows-one_third, stride_length)
count = 0
for frame_idx in tqdm(row_indices, desc="Calculating magnitude distribution.", unit='frames', total=len(row_indices)):
# for count, frame_idx in enumerate(np.arange(one_third,num_rows-one_third, stride_length)):
magnitude, angle = cv2.cartToPolar(flow[frame_idx,:,:,0], flow[frame_idx,:,:,1]);
append_to_mag_histogram(count, magnitude)
count = count+1
plt.close('all')
self.save_out_mag_histogram(algorithm)
def filter_magnitude(self, magnitude, bgr_frame):
if self.cuda_enabled:
_, mask = cv2.cuda.threshold(self.current_gpu, 50, 255, cv2.THRESH_TOZERO)
magnitude = cv2.cuda.bitwise_and(magnitude, magnitude, mask=mask.download())
else:
_, mask = cv2.threshold(self.current_gray, 50, 255, cv2.THRESH_TOZERO)
magnitude = cv2.bitwise_and(magnitude, magnitude, mask=mask)
return magnitude
def apply_magnitude_thresholds_and_rescale(self, magnitude, lower_mag_threshold=False, upper_mag_threshold=False):
if lower_mag_threshold:
magnitude[magnitude < lower_mag_threshold] = 0
if upper_mag_threshold:
magnitude[magnitude > upper_mag_threshold] = upper_mag_threshold
magnitude = (magnitude / upper_mag_threshold) * 255.0
return magnitude
def convert_flow_to_magnitude_angle(self,flow,
bgr_world_in,
lower_mag_threshold = False,
upper_mag_threshold = False):
magnitude, angle = cv2.cartToPolar(flow[...,0], flow[...,1])
# Clips to lower-upper thresh and then rescales into range 0-255
magnitude = self.apply_magnitude_thresholds_and_rescale(magnitude,
lower_mag_threshold=lower_mag_threshold,
upper_mag_threshold=upper_mag_threshold)
# A custom filter. Mine masks out black patches in the input video.
# Due to compression, they can be filled with noise.
processed_bgr_world = self.preprocess_frame(bgr_world_in)
processed_gray_world = cv2.cvtColor(processed_bgr_world, cv2.COLOR_BGR2GRAY)
_, mask = cv2.threshold(processed_gray_world, 10, 255, cv2.THRESH_TOZERO)
magnitude = cv2.bitwise_and(magnitude, magnitude, mask=mask)
angle = cv2.bitwise_and(angle, angle, mask=mask)
return magnitude, angle
@staticmethod
def filter_magnitude(magnitude, bgr_world):
processed_bgr_world = source.preprocess_frame(bgr_world)
processed_gray_world = cv2.cvtColor(processed_bgr_world, cv2.COLOR_BGR2GRAY)
_, mask = cv2.threshold(processed_gray_world, 10, 255, cv2.THRESH_TOZERO)
magnitude = cv2.bitwise_and(magnitude, magnitude, mask=mask)
return magnitude
@staticmethod
def preprocess_frame(frame):
def create_sky_mask_on_hsv(hsv_in):
lower = np.array([100, 45, 100])
upper = np.array([120, 140, 260])
mask = cv2.inRange(hsv_in, lower, upper)
mask = 255 - mask
return mask
def create_road_mask_on_hsv(hsv_in):
lower = np.array([0, 240, 200])
upper = np.array([40, 255, 255])
mask1 = cv2.inRange(hsv, lower, upper)
lower = np.array([170, 200, 100])
upper = np.array([180, 255, 255])
mask2 = cv2.inRange(hsv, lower, upper)
mask = mask1 + mask2
mask = np.clip(mask, 0, 255).astype(np.uint8)
mask = cv2.dilate(mask, kernel=np.ones((5, 5), np.uint8), iterations=2)
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel=np.ones((5, 5), np.uint8))
mask = 255 - mask
return mask
def remove_noise_bgr_dark_patches(bgr_image):
_, mask = cv2.threshold(cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY), 30, 250, cv2.THRESH_TOZERO)
bgr_image = cv2.bitwise_and(bgr_image, bgr_image, mask=mask)
return bgr_image
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
frame = cv2.bitwise_and(frame, frame, mask=create_sky_mask_on_hsv(hsv))
frame = cv2.bitwise_and(frame, frame, mask=create_road_mask_on_hsv(hsv))
frame = remove_noise_bgr_dark_patches(frame)
return frame
def create_video_objects(self,algorithm, visualize_as, gaze_centered, video_out_filename):
video_in = cv2.VideoCapture(self.file_path)
width = int(video_in.get(cv2.CAP_PROP_FRAME_WIDTH))
# If a stacked video viz (e.g., hsv_stacked), double the height
if 'stacked' in visualize_as:
height = int(video_in.get(cv2.CAP_PROP_FRAME_HEIGHT)) * 2
else:
height = int(video_in.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = video_in.get(cv2.CAP_PROP_FPS)
video_out = cv2.VideoWriter(os.path.join(self.video_out_path, video_out_filename),cv2.VideoWriter_fourcc(*'mp4v'), fps, (width,height))
return video_in, video_out
def create_visualization(self,
algorithm: str,
gaze_centered: False,
visualize_as: str,
lower_mag_threshold = False,
upper_mag_threshold = False,
):
if isinstance(self, pupil_labs_source) == False and gaze_centered == True:
logger.error('You can only perform gaze centered operations when using pupil_labs_sources.', stack_info=True, exc_info=True)
sys.exit(1)
# Create video out filename
video_out_filename = []
lower_thresh_string = 'None' if not lower_mag_threshold else str(lower_mag_threshold)
upper_thresh_string = 'None' if not upper_mag_threshold else str(upper_mag_threshold)
thresh_string = f'{lower_thresh_string}-{upper_thresh_string}'
if gaze_centered:
path, input_video_filename = os.path.split(self.file_path)
input_video_filename = input_video_filename.split('.')[0]
video_out_filename = f'{input_video_filename}_gaze_centered_{algorithm}_{thresh_string}.mp4'
else:
video_out_filename = f'{self.source_file_name}_{algorithm}_{thresh_string}.mp4'
# create video containers and streams
video_in, video_out = self.create_video_objects(algorithm, visualize_as, gaze_centered, video_out_filename)
hdf5_object = self.open_hdf5_file(algorithm, gaze_centered)
flow_ds = hdf5_object['flow']
attrs = flow_ds.attrs
width = np.shape(flow_ds[2])
height = np.shape(flow_ds[1])
fps = attrs['approx_fps']
num_frames = int(video_in.get(cv2.CAP_PROP_FRAME_COUNT))
assert np.shape(flow_ds)[0] == attrs['num_source_frames'],"The number of frames from the source does not equal the number of frames in the hdf5 flow file."
for index in tqdm(range(num_frames), desc="Generating " + video_out_filename, unit='frames', total=num_frames):
success, bgr_world = video_in.read()
if not success:
print(f'Frame {index}: video_in.read() unsuccessful')
continue
if gaze_centered:
bgr_world = self.recenter_frame_on_gaze(bgr_world, index)
flow_frame = np.array(flow_ds[index,...])
magnitude, angle = self.convert_flow_to_magnitude_angle(flow_frame,bgr_world,
lower_mag_threshold=lower_mag_threshold,
upper_mag_threshold=upper_mag_threshold)
image_out = self.visualize_frame(index,
bgr_world,
flow_frame,
magnitude,
angle,
visualize_as)
# Add packet to video
video_out.write(image_out)
video_out.release()
video_in.release()
def visualize_frame(self,
index,
current_bgr,
flow,
magnitude,
angle,
visualize_as):
if visualize_as == "streamlines":
image_out = self.visualize_flow_as_streamlines(current_bgr, flow)
elif visualize_as == "vectors":
image_out = self.visualize_flow_as_vectors(current_bgr, magnitude, angle)
elif visualize_as == "hsv_overlay" or visualize_as == "hsv_stacked":
image_out = self.visualize_flow_as_hsv(magnitude, angle)
if visualize_as == "hsv_stacked":
image_out = np.concatenate((current_bgr, image_out), axis=0)
# Save midpoint images?
if self.save_midpoint_images:
if os.path.isdir(self.mid_frames_out_path) is False:
os.makedirs(self.mid_frames_out_path)
cv2.imwrite(str(os.path.join(self.mid_frames_out_path, '{:06d}.png'.format(index))),
image_out, [cv2.IMWRITE_PNG_COMPRESSION, 0])
else:
logger.error('Visualization method not implemented, or a typo', stack_info=True, exc_info=True)
sys.exit(1)
return image_out
def visualize_flow_as_streamlines(self, frame, flow):
x = np.arange(0, np.shape(frame)[0], 1)
y = np.arange(0, np.shape(frame)[1], 1)
grid_x, grid_y = np.meshgrid(y, x)
start_grid_res = 10
start_pts = np.array(
[[y, x] for x in np.arange(0, np.shape(frame)[0], start_grid_res)
for y in np.arange(0, np.shape(frame)[1], start_grid_res)])
fig, ax = plt.subplots(figsize=(6.4, 4.8), dpi=100)
plt.streamplot(grid_x, grid_y, -flow[...,0], -flow[...,1],
start_points=start_pts,
color='w',
maxlength=.4,
arrowsize=0,
linewidth=.8) # density = 1
plt.axis('off')
plt.imshow(frame)
canvas = FigureCanvas(fig)
canvas.draw() # draw the canvas, cache the renderer
width, height = fig.get_size_inches() * fig.get_dpi()
image = np.frombuffer(canvas.tostring_rgb(), dtype=np.uint8).reshape(int(height), int(width), 3)
plt.close('all')
return image
def visualize_flow_as_hsv(self, magnitude, angle):
'''
Note that to perform well, this function really needs an upper_bound, which also acts as a normalizing term.
'''
# create hsv output for optical flow
hsv = np.zeros([np.shape(magnitude)[0], np.shape(magnitude)[1], 3], np.uint8)
hsv[..., 0] = angle * 180 / np.pi / 2 # angle_rads -> degs 0-360 -> degs 0-180
hsv[..., 1] = 255
hsv[..., 2] = magnitude
# cv2.normalize(magnitude, None, 0, 255, cv2.NORM_MINMAX)
hsv_8u = np.uint8(hsv)
bgr = cv2.cvtColor(hsv_8u, cv2.COLOR_HSV2BGR)
return bgr
def visualize_flow_as_vectors(self, image_in, magnitude, angle, skippts=15, scale=.1,scale_units='width',
width=.003, return_image=True):
dpi = 100
fig, ax = plt.subplots()
canvas = FigureCanvas(fig)
ax.margins(0)
s = slice(None, None, skippts)
xmax = np.shape(image_in)[1]
xpoints = int(np.shape(image_in)[1])
x = np.linspace(0, np.shape(image_in)[1], xmax)
ymax = np.shape(image_in)[0]
ypoints = int(np.shape(image_in)[0])
y = np.linspace(0, np.shape(image_in)[0], ymax)
x = x[s]
y = y[s]
x2d, y2d = np.meshgrid(x, y, indexing='xy')
u = magnitude * np.cos(angle) * -1
v = magnitude * np.sin(angle)
u = (u[s, s] / 255) * scale
v = (v[s, s] / 255) * scale
plt.imshow(cv2.cvtColor(image_in, cv2.COLOR_BGR2RGB))
ax.axis('off')
plt.quiver(x2d, y2d, u, v, color='white', alpha=0.7, width=width, scale=scale,
scale_units='inches')
#if return_image:
canvas.draw() # draw the canvas, cache the renderer
image_from_plot = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
image_from_plot = image_from_plot.reshape(fig.canvas.get_width_height()[::-1] + (3,))
plt.close('all')
return cv2.cvtColor(image_from_plot,cv2.COLOR_BGR2RGB)
# def visualize_flow_as_vectors(self, frame, magnitude, angle, divisor=15):
#
# '''Display image with a visualisation of a flow over the top.
# A divisor controls the density of the quiver plot.'''
#
# # create a blank mask, on which lines will be drawn.
# mask = np.zeros([np.shape(magnitude)[0], np.shape(magnitude)[1], 3], np.uint8)
#
# # if vector_scalar != 1 & vector_scalar != False:
# # magnitude = np.multiply(magnitude, vector_scalar)
#
# vector_x, vector_y = cv2.polarToCart(magnitude, angle)
#
# for r in range(1, int(np.shape(magnitude)[0] / divisor)):
# for c in range(1, int(np.shape(magnitude)[1] / divisor)):
# origin_x = c * divisor
# origin_y = r * divisor
#
# endpoint_x = int(origin_x + vector_x[origin_y, origin_x])
# endpoint_y = int(origin_y + vector_y[origin_y, origin_x])
#
# mask = cv2.arrowedLine(mask, (origin_x, origin_y), (endpoint_x, endpoint_y), color=(0, 0, 255),
# thickness=3, tipLength=0.35)
#
# return cv2.addWeighted(frame, 0.5, mask, 0.5, 0)
def set_video_target(self, initial_dir=False, file_path=False):
if not file_path:
if initial_dir:
Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
file_path = askopenfilename(title="Select the video target",initialdir=initial_dir)
else:
title="Select the video target"
Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
file_path = askopenfilename(title="Select the video target", initialdir=self.video_out_path)
self.video_target_path = file_path
return True
def avg_flow_magnitude_by_direction(self,start_frame=False,play_video=False,save_video=False):
from matplotlib import cm
from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas
if self.video_target_path == False:
self.set_video_target()
video_in = cv2.VideoCapture(self.video_target_path)
width = int(video_in.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(video_in.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = video_in.get(cv2.CAP_PROP_FPS)
if save_video:
path, input_video_filename = os.path.split(self.video_target_path)
input_video_filename = input_video_filename.split('.')[0]
video_out_name = f'{input_video_filename}_avg-mag-by-direction.mp4'
video_out = cv2.VideoWriter(os.path.join(self.export_folder, video_out_name),
cv2.VideoWriter_fourcc(*'mp4v'), fps, (width, 2*height))
# from pathlib import Path
# p = Path(
# #'D:/Github/retinal_flow_toolkit/pupil_labs_data/cb13/S001/PupilData/000/exports/000/cb13_world_nvidia2_hsv_overlay.mp4'
# 'D:\Github\retinal_flow_toolkit\demo_input_video\dash_cam.mp4'
# ).as_posix()
# p = Path('/Users/gjdpci/Documents/Data/Steering/cb13/S001/PupilData/000/exports/000/cb13_world_nvidia2_hsv_overlay.mp4 ')
# self.video_target_path = p.as_posix()
# video = cv2.VideoCapture('cb13_world_nvidia2_hsv_overlay.mp4')
count = 0
success = 1
hist_params = (4 *2, 0, 360)
bins = np.linspace(hist_params[1], hist_params[2], hist_params[0] + 1)
bin_centers = [(bins[i] + bins[i + 1]) / 2 for i in range(len(bins) - 1)]
cvals = np.array(bin_centers) / 359 + 0.5
cvals[cvals > 1] = cvals[cvals > 1] - 1
hsv_map = cm.get_cmap('hsv').reversed()
bar_colors = hsv_map(cvals)
from collections import deque
bin_rad_hist = deque(maxlen=30)
# start_frame = 16200 # right turn, high flow
#start_frame = 17800 # lef turn, high flow
# start_frame = 15725 # right turn, med flo
# start_frame = 30000
num_frames = int(video_in.get(cv2.CAP_PROP_FRAME_COUNT))
if start_frame:
video_in.set(cv2.CAP_PROP_POS_FRAMES, start_frame)
else:
start_frame= 0
for index in tqdm(range(num_frames-start_frame), desc="Calculating avg vector magnitude", unit='frames', total=num_frames-start_frame):
success, image = video_in.read()
if success:
dpi = 100
fig, ax = plt.subplots(figsize=(width / dpi, height / dpi), subplot_kw=dict(projection='polar'))
canvas = FigureCanvas(fig)
ax.margins(0)
ax.set_ylim([0, 200])
hsv_image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
h = hsv_image[..., 0]
v = hsv_image[..., 2]
mag_flat = v.flatten()[v.flatten() > 10]
hue_flat = h.flatten()[v.flatten() > 10]
hue_flat = hue_flat * 2.0
bin_rad = []
for b_idx in range(1, len(bins)):
idx = np.where(np.logical_and(hue_flat >= bins[b_idx - 1], hue_flat <= bins[b_idx]))
bin_rad.append(np.mean(mag_flat[idx]))
bin_rad = np.nan_to_num(bin_rad)
bin_rad_hist.appendleft(bin_rad)
ax.bar(np.deg2rad(bin_centers), np.mean(bin_rad_hist,axis=0), width=2 * (np.pi / (len(bins))), color=bar_colors);
canvas.draw() # draw the canvas, cache the renderer
image_from_plot = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
image_from_plot = image_from_plot.reshape(fig.canvas.get_width_height()[::-1] + (3,))
combined_image = np.vstack([image, cv2.cvtColor(image_from_plot, cv2.COLOR_RGB2BGR)])
plt.close('all')
if save_video:
video_out.write(combined_image)
if play_video:
cv2.imshow(f'Image', combined_image)
key = cv2.waitKey(25) # pauses for N ms before fetching next image
if key == 27: # if ESC is pressed, exit loop
if save_video:
video_out.release()
video_in.release()
cv2.destroyAllWindows()
return
cv2.destroyAllWindows()
video_out.release()
video_in.release()
def recenter_frame_on_gaze(self, frame, frame_index):
logger.error('Gaze centering is only possible with pupil labs data sources',
stack_info=True, exc_info=True)
sys.exit(1)
class pupil_labs_source(video_source):
def __init__(self, pupil_labs_parent_folder,
session_number=False,
recording_number=False,
export_number=False,
analysis_parameters_file='default_analysis_parameters.json',
recalculate_processed_gaze = False):
self.analysis_parameters = json.load(open(analysis_parameters_file))
self.pupil_labs_parent_folder = pupil_labs_parent_folder
self.session_number = session_number
self.recording_number = recording_number
self.export_number = export_number
self.recording_folder = self.set_recording_folder()
self.export_folder = self.set_export_folder()
self.world_video_path = os.path.join(self.recording_folder, 'world.mp4')
self.gaze_overlay_video_out_name = False
self.gaze_data = False
self.processed_gaze_data = False
# Load processed gaze data from file
proc_gaze_file_path = os.path.join(self.export_folder, 'processed_gaze.pkl')
if recalculate_processed_gaze is False and os.path.exists(proc_gaze_file_path):
with open(proc_gaze_file_path, 'rb') as handle:
self.processed_gaze_data = pickle.load(handle)
super().__init__(self.world_video_path, out_parent_dir=self.export_folder)
self.raw_frames_out_path = os.path.join(self.out_parent_dir, 'world_images')
self.mid_frames_out_path = os.path.join(self.out_parent_dir, 'flow_mid_images')
self.flow_frames_out_path = os.path.join(self.out_parent_dir, 'flow_images')
self.video_out_path = self.out_parent_dir
self.magnitude_out_path = os.path.join(self.out_parent_dir, 'flow_magnitude_data')
f = open(os.path.join(self.recording_folder, 'info.player.json'))
self.player_info = json.load(f)
f.close()
def calculate_flow(self,
video_out_name=False,
algorithm="nvidia2",
preprocess_frames=True,
gaze_centered=False,
save_input_images=False,
save_output_images=False):
if gaze_centered:
self.import_gaze_from_exports()
self.process_gaze_data()
super().calculate_flow(video_out_name=video_out_name,
algorithm=algorithm,
preprocess_frames=preprocess_frames,
gaze_centered=gaze_centered,
save_input_images=save_input_images,
save_output_images=save_output_images)
def modify_frame(self, frame, frame_index):
self.gaze_center_Frame(frame, frame_index)
def get_median_gaze_for_frame(self, frame_index):