-
Notifications
You must be signed in to change notification settings - Fork 16
/
Copy pathtest-surveyed-calibration.py
executable file
·847 lines (685 loc) · 34 KB
/
test-surveyed-calibration.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
#!/usr/bin/env python3
r'''Surveyed camera-calibration test
I observe, with noise, a number of fixed, known-pose chessboards from a single
camera. And I make sure that I can more or less compute the camera intrinsics
and extrinsics.
This is a monocular solve. Since the chessboards aren't allowed to move,
multiple cameras in one solves wouldn't be interacting with each other, and
there's no reason to have more than one
'''
import sys
import argparse
import re
import os
def parse_args():
parser = \
argparse.ArgumentParser(description = __doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument('--write-model',
type=str,
help='''If given, we write the resulting model to disk
for further analysis. The filename is given in this
argument''')
parser.add_argument('--range-board',
type=float,
default = 10,
help='''Distance from the camera to the center of each
chessboard. If omitted, we default to 10''')
parser.add_argument('--range-board-center',
type=float,
help='''Distance from the camera to the center of the
MIDDLE chessboard. If omitted, we use --range-board * 2.
Everything at one range causes poor calibrations, and I
don't want that to be the default case here''')
parser.add_argument('--oversample',
type=int,
default = 1,
help='''How many times we observe the stationary scene.
Observing multiple times gives us more samples of the
noise, resulting in lower uncertainties''')
parser.add_argument('--distance',
type=float,
help='''distance for uncertainty computations. If
omitted, we use infinity''')
parser.add_argument('--seed-rng',
type=int,
default=0,
help='''Value to seed the rng with''')
parser.add_argument('--do-sample',
action='store_true',
help='''By default we don't run the time-intensive
samples of the calibration solves. This runs a very
limited set of tests, and exits. To perform the full set
of tests, pass --do-sample''')
parser.add_argument('--only-report-uncertainty',
action='store_true',
help='''If given, I ONLY report the uncertainty in projection and errz''')
parser.add_argument('--Nsamples',
type=int,
default=500,
help='''How many random samples to evaluate''')
parser.add_argument('--make-documentation-plots',
type=str,
help='''If given, we produce plots for the
documentation. Takes one argument: a string describing
this test. This will be used in the filenames of the
resulting plots. To make interactive plots, pass ""''')
parser.add_argument('--terminal-pdf',
type=str,
help='''The gnuplotlib terminal for --make-documentation-plots .PDFs. Omit this
unless you know what you're doing''')
parser.add_argument('--terminal-svg',
type=str,
help='''The gnuplotlib terminal for --make-documentation-plots .SVGs. Omit this
unless you know what you're doing''')
parser.add_argument('--terminal-png',
type=str,
help='''The gnuplotlib terminal for --make-documentation-plots .PNGs. Omit this
unless you know what you're doing''')
args = parser.parse_args()
if args.range_board_center is None:
args.range_board_center = args.range_board * 2.
return args
args = parse_args()
import numpy as np
import numpysane as nps
import os
testdir = os.path.dirname(os.path.realpath(__file__))
# I import the LOCAL mrcal since that's what I'm testing
sys.path[:0] = f"{testdir}/..",
import mrcal
import testutils
from test_calibration_helpers import sample_dqref,calibration_sample
import copy
# I Reduce FOV to make it clear that you need data from different ranges
focal_length_scale_from_true = 10.
object_spacing = 0.03
object_width_n = 8
object_height_n = 8
pixel_uncertainty_stdev = 0.5
# I have a 3x3 grid of chessboards
NboardgridX,NboardgridY = 3,1
# We try to recover this in the calibration process
# shape (6,)
rt_cam_ref_true = np.array((-0.04, 0.05, -0.1, 1.2, -0.1, 0.1),)
random_radius__r_cam_board_true = 0.1
random_radius__t_cam_board_true = 1.0e-1
############# Plot setup. Very similar to test-projection-uncertainty.py
if args.make_documentation_plots is not None:
def shorter_terminal(t):
# Adjust the terminal string to be less tall. Reduces wasted space in
# some of the plots
m = re.match("(.*)( size.*?,)([0-9.]+)(.*?)$", t)
if m is None: return t
return m.group(1) + m.group(2) + str(float(m.group(3))*0.8) + m.group(4)
import gnuplotlib as gp
terminal = dict(pdf = args.terminal_pdf,
svg = args.terminal_svg,
png = args.terminal_png,
gp = 'gp')
pointscale = dict(pdf = 1,
svg = 1,
png = 1,
gp = 1)
pointscale[""] = 1.
if args.make_documentation_plots:
print(f"Will write documentation plots to {args.make_documentation_plots}-xxxx.pdf and .svg")
if terminal['svg'] is None: terminal['svg'] = 'svg size 800,600 noenhanced solid dynamic font ",14"'
if terminal['pdf'] is None: terminal['pdf'] = 'pdf size 8in,6in noenhanced solid color font ",12"'
if terminal['png'] is None: terminal['png'] = 'pngcairo size 1024,768 transparent noenhanced crop font ",12"'
extraset = dict()
for k in pointscale.keys():
extraset[k] = f'pointsize {pointscale[k]}'
# I want the RNG to be deterministic
np.random.seed(args.seed_rng)
############# Set up my world, and compute all the perfect positions, pixel
############# observations of everything
model_true = mrcal.cameramodel(f"{testdir}/data/cam0.opencv8.cameramodel")
imagersize_true = model_true.imagersize()
W,H = imagersize_true
# I have opencv8 model_true, but I truncate to opencv4 to keep this simple and
# fast
lensmodel = 'LENSMODEL_OPENCV4'
Nintrinsics = mrcal.lensmodel_num_params(lensmodel)
intrinsics_data_true = model_true.intrinsics()[1][:Nintrinsics]
intrinsics_data_true[:2] *= focal_length_scale_from_true
model_true.intrinsics( (lensmodel, intrinsics_data_true), )
model_true.extrinsics_rt_fromref(rt_cam_ref_true)
if False:
model_true.write("/tmp/true.cameramodel")
print("wrote /tmp/true.cameramodel")
# Test point. On the left, centered vertically
q_query = np.array((W/4, H/2))
# I want to space out the chessboards across the whole imager. To make this
# simpler I define it in terms of the "boardcentered" coordinate system, whose
# origin is at the center of the chessboard, not in a corner
board_center = \
np.array(( (object_width_n -1)*object_spacing/2.,
(object_height_n-1)*object_spacing/2.,
0 ))
rt_boardcentered_board_true = mrcal.identity_rt()
rt_boardcentered_board_true[3:] = -board_center
# The chessboard centers are arranged in an even grid on the imager
#
# shape (NboardgridY,NboardgridX,2); each row is (qx,qy): the center of each
# chessboard
q_center = \
nps.mv( \
nps.cat(*np.meshgrid(np.linspace(0.5, NboardgridX-0.5, NboardgridX, endpoint=True)/NboardgridX * W,
np.linspace(0.5, NboardgridY-0.5, NboardgridY, endpoint=True)/NboardgridY * H)),
0, -1)
v_center = mrcal.unproject(np.ascontiguousarray(q_center),
*model_true.intrinsics(),
normalize = True)
# shape (NboardgridY,NboardgridX,6)
rt_cam_boardcentered_true = np.zeros(v_center.shape[:-1] + (6,), dtype=float)
rt_cam_boardcentered_true[...,3:] = v_center*args.range_board
if args.range_board_center is not None:
rt_cam_boardcentered_true[NboardgridY//2, NboardgridX//2,3:] = \
v_center[NboardgridY//2, NboardgridX//2,:]*args.range_board_center
# shape (NboardgridY,NboardgridX,6)
rt_cam_board_true = mrcal.compose_rt(rt_cam_boardcentered_true,
rt_boardcentered_board_true)
rt_cam_board_true[...,:3] += (np.random.rand(*rt_cam_board_true[...,:3].shape)*2. - 1.) * random_radius__r_cam_board_true
rt_cam_board_true[...,3:] += (np.random.rand(*rt_cam_board_true[...,3:].shape)*2. - 1.) * random_radius__t_cam_board_true
# Nboards = NboardgridX*NboardgridY
# shape (Nboards,6)
rt_cam_board_true = nps.clump(rt_cam_board_true, n=2)
rt_ref_board_true = mrcal.compose_rt( mrcal.invert_rt(rt_cam_ref_true),
rt_cam_board_true)
# We assume that the chessboard poses rt_ref_board_true, rt_cam_board_true were
# measured. Perfectly. This it the ground truth AND we have it available in the
# calibration
# shape (Nh,Nw,3)
cal_object = mrcal.ref_calibration_object(object_width_n,
object_height_n,
object_spacing)
# shape (Nframes,Nh,Nw,3)
pcam_calobjects = \
mrcal.transform_point_rt(nps.mv(rt_cam_board_true,-2,-4), cal_object)
# shape (N,3)
pcam_true = nps.clump(pcam_calobjects, n=3)
pref_true = mrcal.transform_point_rt(mrcal.invert_rt(rt_cam_ref_true),
pcam_true)
# shape (N,2)
q_true = mrcal.project(pcam_true, lensmodel, intrinsics_data_true)
Npoints = q_true.shape[0]
if False:
# show the angles off the optical axis
import gnuplotlib as gp
p = pcam_true / nps.dummy(nps.mag(pcam_true),-1)
c = p[...,2]
th = np.arccos(c).ravel() * 180./np.pi
gp.plot(th, yrange=(45,90))
import IPython
IPython.embed()
sys.exit()
# I have perfect observations in q_true. I corrupt them by noise weight has
# shape (N,),
weight01 = (np.random.rand(*q_true.shape[:-1]) + 1.) / 2. # in [0,1]
weight0 = 0.2
weight1 = 1.0
weight = weight0 + (weight1-weight0)*weight01
# out-of-bounds observations are outliers
weight[(q_true[:,0] < 0) + \
(q_true[:,1] < 0) + \
(q_true[:,0] > W-1) + \
(q_true[:,1] > H-1)] = -1.
############# Now I pretend that the noisy observations are all I got, and I run
############# a calibration from those
# shape (Noversample, N, 3)
observations_point = np.zeros((args.oversample,Npoints,3), dtype=float)
# write the same perfect data into each oversampled slice
observations_point += nps.glue(q_true,
nps.dummy(weight,-1),
axis=-1)
# shape (Noversample*N, 3)
observations_point = nps.clump(observations_point,n=2)
# Dense observations. The cameras see all the boards
indices_point_camintrinsics_camextrinsics = np.zeros( (args.oversample,Npoints, 3), dtype=np.int32)
indices_point_camintrinsics_camextrinsics[...,0] += np.arange(Npoints)
indices_point_camintrinsics_camextrinsics = nps.clump(indices_point_camintrinsics_camextrinsics,n=2)
if False:
# Compute the seed using the seeding algorithm
focal_estimate = 2000 * focal_length_scale_from_true
cxy = np.array(( (W-1.)/2,
(H-1.)/2, ), )
intrinsics_core_estimate = \
('LENSMODEL_STEREOGRAPHIC',
np.array((focal_estimate,
focal_estimate,
*cxy
)))
# I select data just at the center for seeding. Eventually should move this to
# the general logic in calibration.py
i = nps.norm2( observations_point[:Npoints,:2] - cxy ) < 1000**2.
observations_point_center = np.array(observations_point[:Npoints])
observations_point_center[~i,2] = -1.
Rt_camera_ref_estimate = \
mrcal.calibration._estimate_camera_pose_from_point_observations( \
indices_point_camintrinsics_camextrinsics[:Npoints,:],
observations_point_center,
(intrinsics_core_estimate,),
pref_true,
icam_intrinsics = 0)
rt_camera_ref_estimate = mrcal.rt_from_Rt(Rt_camera_ref_estimate)
intrinsics_data = np.zeros((1,Nintrinsics), dtype=float)
intrinsics_data[:,:4] = intrinsics_core_estimate[1]
intrinsics_data[:,4:] = np.random.random( (1, intrinsics_data.shape[1]-4) ) * 1e-6
else:
# Use a nearby value to the true seed
rt_camera_ref_estimate = nps.atleast_dims(np.array(rt_cam_ref_true), -2)
rt_camera_ref_estimate[..., :3] += np.random.randn(3) * 1e-2 # r
rt_camera_ref_estimate[..., 3:] += np.random.randn(3) * 0.1 # t
intrinsics_data = nps.atleast_dims(np.array(intrinsics_data_true), -2)
intrinsics_data += np.random.randn(*intrinsics_data.shape) * 1e-2
if args.make_documentation_plots is not None:
def makeplot(**plotoptions):
mrcal.show_geometry( (rt_camera_ref_estimate,),
points = pref_true, # known. fixed. perfect.
show_points = True,
title = 'Surveyed calibration: nominal geometry',
**plotoptions)
if args.make_documentation_plots:
for extension in ('pdf','svg','png','gp'):
makeplot(wait = False,
terminal = shorter_terminal(terminal[extension]),
_set = extraset[extension],
hardcopy = f'{args.make_documentation_plots}/geometry.{extension}')
else:
makeplot(wait = True)
if args.make_documentation_plots is not None:
def makeplot(**plotoptions):
gp.plot( observations_point[:,:2],
_with = f'points ps {pointscale[extension]}',
tuplesize = -2,
square=1,
_xrange = f'0:{W-1}',
_yrange = f'{H-1}:0',
title = 'Surveyed calibration: observations',
**plotoptions)
if args.make_documentation_plots:
for extension in ('pdf','svg','png','gp'):
makeplot(wait = False,
terminal = terminal[extension],
_set = extraset[extension],
hardcopy = f'{args.make_documentation_plots}/observations.{extension}')
else:
makeplot(wait = True)
optimization_inputs_baseline = \
dict( lensmodel = lensmodel,
intrinsics = intrinsics_data,
extrinsics_rt_fromref = rt_camera_ref_estimate,
frames_rt_toref = None,
points = pref_true, # known. fixed. perfect.
observations_board = None,
indices_frame_camintrinsics_camextrinsics = None,
observations_point = observations_point,
indices_point_camintrinsics_camextrinsics = indices_point_camintrinsics_camextrinsics,
calobject_warp = None,
imagersizes = nps.atleast_dims(imagersize_true, -2),
point_min_range = 0.01,
point_max_range = 100.,
verbose = False,
do_apply_outlier_rejection = False,
do_apply_regularization = True,
do_optimize_calobject_warp = False,
do_optimize_frames = False)
Nstate = mrcal.num_states(**optimization_inputs_baseline)
def optimize(optimization_inputs):
optimization_inputs['do_optimize_intrinsics_core'] = False
optimization_inputs['do_optimize_intrinsics_distortions'] = False
optimization_inputs['do_optimize_extrinsics'] = True
mrcal.optimize(**optimization_inputs)
optimization_inputs['do_optimize_intrinsics_core'] = True
optimization_inputs['do_optimize_intrinsics_distortions'] = True
optimization_inputs['do_optimize_extrinsics'] = True
mrcal.optimize(**optimization_inputs)
# Take one solve sample
optimization_inputs = copy.deepcopy(optimization_inputs_baseline)
_,optimization_inputs['observations_point'] = \
sample_dqref(optimization_inputs['observations_point'], pixel_uncertainty_stdev)
optimize(optimization_inputs)
# Grab the measurements. measurements_point() reports ONLY the point
# reprojection errors: no board observations, no regularization. Also, no
# outliers
rmserr_point = np.std(mrcal.measurements_point(optimization_inputs).ravel())
############# Calibration computed. Now I see how well I did
model_solved = \
mrcal.cameramodel( optimization_inputs = optimization_inputs,
icam_intrinsics = 0 )
Rt_extrinsics_err = \
mrcal.compose_Rt( model_solved.extrinsics_Rt_fromref(),
model_true .extrinsics_Rt_toref() )
v_query_cam = mrcal.unproject(q_query, *model_solved.intrinsics(), normalize = True)
if args.distance is not None:
p_query_cam = v_query_cam * args.distance
else:
p_query_cam = v_query_cam * 1e5 # large distance
p_query_ref = mrcal.transform_point_Rt(model_solved.extrinsics_Rt_toref(),p_query_cam)
def get_variances():
icam = 0
i_state = mrcal.state_index_extrinsics(icam, **optimization_inputs)
rt_extrinsics_err, \
drt_extrinsics_err_drtsolved, \
_ = \
mrcal.compose_rt( model_solved.extrinsics_rt_fromref(),
model_solved.extrinsics_rt_toref(),
get_gradients = True)
derrz_db = np.zeros( (1,Nstate), dtype=float)
derrz_db[...,i_state:i_state+6] = drt_extrinsics_err_drtsolved[5:,:]
Var_errz = mrcal.model_analysis._propagate_calibration_uncertainty( \
'covariance',
dF_db = derrz_db,
optimization_inputs = optimization_inputs)
Var_errz = Var_errz[0,0]
if args.distance is not None:
Var_q = \
mrcal.projection_uncertainty( p_query_cam, model_solved,
atinfinity = False,
what = 'covariance' )
else:
Var_q = \
mrcal.projection_uncertainty( v_query_cam, model_solved,
atinfinity = True,
what = 'covariance' )
return Var_errz,Var_q
if args.only_report_uncertainty:
Var_errz,Var_q = get_variances()
print("# z-bulk z-center stdev(q) stdev(errz)")
# using the 'rms-stdev' expression from _propagate_calibration_uncertainty()
print(f"{args.range_board} {args.range_board_center} {np.sqrt(nps.trace(Var_q)/2.):.4f} {np.sqrt(Var_errz):.4f}")
sys.exit()
# verify problem layout
testutils.confirm_equal( Nstate,
Nintrinsics + 6,
msg="num_states()")
testutils.confirm_equal( mrcal.num_states_intrinsics(**optimization_inputs),
Nintrinsics,
msg="num_states_intrinsics()")
testutils.confirm_equal( mrcal.num_intrinsics_optimization_params(**optimization_inputs),
Nintrinsics,
msg="num_intrinsics_optimization_params()")
testutils.confirm_equal( mrcal.num_states_extrinsics(**optimization_inputs),
6,
msg="num_states_extrinsics()")
testutils.confirm_equal( mrcal.num_states_frames(**optimization_inputs),
0,
msg="num_states_frames()")
testutils.confirm_equal( mrcal.num_states_points(**optimization_inputs),
0,
msg="num_states_points()")
testutils.confirm_equal( mrcal.num_states_calobject_warp(**optimization_inputs),
0,
msg="num_states_calobject_warp()")
testutils.confirm_equal( mrcal.num_measurements_boards(**optimization_inputs),
0,
msg="num_measurements_boards()")
testutils.confirm_equal( mrcal.num_measurements_points(**optimization_inputs),
Npoints*args.oversample*2,
msg="num_measurements_points()")
testutils.confirm_equal( mrcal.num_measurements_regularization(**optimization_inputs),
6,
msg="num_measurements_regularization()")
testutils.confirm_equal( mrcal.state_index_intrinsics(0, **optimization_inputs),
0,
msg="state_index_intrinsics()")
testutils.confirm_equal( mrcal.state_index_extrinsics(0, **optimization_inputs),
8,
msg="state_index_extrinsics()")
testutils.confirm_equal( mrcal.measurement_index_points(2, **optimization_inputs),
2*2,
msg="measurement_index_points()")
testutils.confirm_equal( mrcal.measurement_index_regularization(**optimization_inputs),
2*args.oversample*Npoints,
msg="measurement_index_regularization()")
if args.write_model:
model_solved.write(args.write_model)
print(f"Wrote '{args.write_model}'")
testutils.confirm_equal(rmserr_point, 0,
eps = 2.5,
msg = "Converged to a low RMS error")
# I expect the fitted error (rmserr_point) to be a bit lower than
# pixel_uncertainty_stdev because this solve is going to overfit: I don't have
# enough data to uniquely define this model. And this isn't even a flexible
# splined model!
testutils.confirm_equal( rmserr_point,
pixel_uncertainty_stdev - pixel_uncertainty_stdev*0.12,
eps = pixel_uncertainty_stdev*0.12,
msg = "Residual have the expected distribution" )
testutils.confirm_equal( nps.mag(Rt_extrinsics_err[3,:]),
0.0,
eps = 0.05,
msg = "Recovered extrinsic translation")
testutils.confirm_equal( (np.trace(Rt_extrinsics_err[:3,:]) - 1) / 2.,
1.0,
eps = np.cos(1. * np.pi/180.0), # 1 deg
msg = "Recovered extrinsic rotation")
# Checking the intrinsics. Each intrinsics vector encodes an implicit
# transformation. I compute and apply this transformation when making my
# intrinsics comparisons. I make sure that within some distance of the pixel
# center, the projections match up to within some number of pixels
Nw = 60
def projection_diff(models, max_dist_from_center):
lensmodels = [model.intrinsics()[0] for model in models]
intrinsics_data = [model.intrinsics()[1] for model in models]
imagersizes = [model.imagersize() for model in models]
# v shape (...,Ncameras,Nheight,Nwidth,...)
# q0 shape (..., Nheight,Nwidth,...)
v,q0 = \
mrcal.sample_imager_unproject(Nw,None,
*imagersizes[0],
lensmodels, intrinsics_data,
normalize = True)
focus_center = None
focus_radius = -1
if focus_center is None: focus_center = ((W-1.)/2., (H-1.)/2.)
if focus_radius < 0: focus_radius = min(W,H)/6.
implied_Rt10 = \
mrcal.implied_Rt10__from_unprojections(q0,
v[0,...], v[1,...],
focus_center = focus_center,
focus_radius = focus_radius)
q1 = mrcal.project( mrcal.transform_point_Rt(implied_Rt10,
v[0,...]),
lensmodels[1], intrinsics_data[1])
diff = nps.mag(q1 - q0)
# zero-out everything too far from the center
center = (imagersizes[0] - 1.) / 2.
diff[ nps.norm2(q0 - center) > max_dist_from_center*max_dist_from_center ] = 0
if False:
import gnuplotlib as gp
gp.plot(diff,
ascii = True,
using = mrcal.imagergrid_using(imagersizes[0], Nw),
square=1, _with='image', tuplesize=3, hardcopy='/tmp/yes.gp', cbmax=3)
return diff
diff = projection_diff( (model_true, model_solved), 800)
testutils.confirm_equal(diff, 0,
worstcase = True,
eps = 6.,
msg = "Recovered intrinsics")
# I test make_perfect_observations(). Doing it here is easy; doing it elsewhere
# it much more work
if True:
optimization_inputs_perfect = copy.deepcopy(optimization_inputs)
mrcal.make_perfect_observations(optimization_inputs_perfect,
observed_pixel_uncertainty=0)
x = mrcal.optimizer_callback(**optimization_inputs_perfect,
no_jacobian = True,
no_factorization = True)[1]
Nmeas = mrcal.num_measurements_points(**optimization_inputs_perfect)
if Nmeas > 0:
i_meas0 = mrcal.measurement_index_points(0, **optimization_inputs_perfect)
testutils.confirm_equal( x[i_meas0:i_meas0+Nmeas],
0,
worstcase = True,
eps = 1e-8,
msg = f"make_perfect_observations() works for points")
else:
testutils.confirm( False,
msg = f"Nmeasurements_points <= 0")
if not args.do_sample:
testutils.finish()
sys.exit()
#### We're going to computed a bunch of noisy calibrations, and gather
#### statistics. I look at:
#### - the projection of one point
#### - the error in the z coordinate of the camera position
####
#### For each one I compute the predicted uncertainty and the empirical one,
#### from the samples, and make sure they match
( intrinsics_sampled, \
rt_cam_ref_sampled_mounted, \
frames_sampled, \
points_sampled, \
calobject_warp_sampled, \
q_noise_board_sampled, \
q_noise_point_sampled, \
b_sampled, \
optimization_inputs_sampled ) = \
calibration_sample(args.Nsamples,
optimization_inputs_baseline,
pixel_uncertainty_stdev,
fixedframes = True,
function_optimize = optimize)
Var_errz,Var_q = get_variances()
####### check errz
if 1:
rt_extrinsics_sampled_err = \
mrcal.compose_rt( rt_cam_ref_sampled_mounted,
model_solved.extrinsics_rt_toref())
errz_sampled = rt_extrinsics_sampled_err[:,0,5:]
testutils.confirm_equal( np.mean(errz_sampled),
0,
eps = 0.01,
msg = f"errz distribution has mean=0")
testutils.confirm_equal( np.std(errz_sampled),
np.sqrt(Var_errz),
eps=0.02,
relative = True,
msg = f"Predicted Var(errz) correct")
if args.make_documentation_plots is not None:
errz_sampled = rt_extrinsics_sampled_err[...,0,5]
binwidth = 0.01
equation_observed = \
mrcal.fitted_gaussian_equation(x = errz_sampled,
binwidth = binwidth,
legend = f'Observed in simulation')
equation_predicted = \
mrcal.fitted_gaussian_equation(mean = np.mean(errz_sampled),
sigma = np.sqrt(Var_errz),
N = len(errz_sampled),
binwidth = binwidth,
legend = "Predicted by mrcal")
def makeplot(**plotoptions):
gp.plot(errz_sampled,
histogram = True,
binwidth = binwidth,
equation_above = (equation_observed,
equation_predicted),
xlabel = "Error in z (m)",
ylabel = "Count",
title = "Surveyed calibration: distribution of error-in-z due to calibration-time noise",
**plotoptions)
if args.make_documentation_plots:
for extension in ('pdf','svg','png','gp'):
makeplot(wait = False,
terminal = terminal[extension],
_set = extraset[extension],
hardcopy = f'{args.make_documentation_plots}/var-errz.{extension}')
else:
makeplot(wait = True)
####### check projection of a point
if 1:
if 1:
# Code check. Computing this directly should mimic the
# mrcal.projection_uncertainty() result exactly since it should be 100% the
# same computation.
#
# EXCEPT when looking at infinity. The projection_uncertainty() path
# uses atinfinity=True, so it ignores translations completely. THIS path
# uses a high distance, so the result will be close, but not identical.
# I use a sloppier bound in that case
p,dp_drt_cam_ref,_ = \
mrcal.transform_point_rt(model_solved.extrinsics_rt_fromref(),
p_query_ref,
get_gradients = True)
q,dq_dp,dq_di = \
mrcal.project( p,
model_solved.intrinsics()[0],
model_solved.intrinsics()[1],
get_gradients = True)
dq_drt_cam_ref = nps.matmult(dq_dp, dp_drt_cam_ref)
dq_db = np.zeros( (2,Nstate), dtype=float)
icam = 0
i_state = mrcal.state_index_extrinsics(icam, **optimization_inputs)
dq_db[...,i_state:i_state+6] = dq_drt_cam_ref
i_state = mrcal.state_index_intrinsics(icam, **optimization_inputs)
dq_db[...,i_state:i_state+Nintrinsics] = dq_di
Var_q2 = mrcal.model_analysis._propagate_calibration_uncertainty( \
'covariance',
dF_db = dq_db,
optimization_inputs = optimization_inputs)
testutils.confirm_equal(Var_q, Var_q2,
relative=True,
worstcase=True,
eps=1e-6 if args.distance is not None else 1e-2,
msg="projection_uncertainty() should do the same thing s caling _propagate_calibration_uncertainty() directly")
p_query_cam_sampled = \
mrcal.transform_point_rt(rt_cam_ref_sampled_mounted[:,0,:],
p_query_ref)
q_query_sampled = \
mrcal.project( p_query_cam_sampled,
model_solved.intrinsics()[0],
intrinsics_sampled[:,0,:] )
q_query_sampled_mean = np.mean(q_query_sampled, axis=0)
# shape (2,2)
Var_q_observed = np.mean( nps.outer(q_query_sampled-q_query_sampled_mean,
q_query_sampled-q_query_sampled_mean),
axis=0 )
worst_direction_stdev_observed = mrcal.worst_direction_stdev(Var_q_observed)
worst_direction_stdev_predicted = mrcal.worst_direction_stdev(Var_q)
testutils.confirm_equal( nps.mag(q_query_sampled_mean - q_query),
0,
eps = 5,
worstcase = True,
msg = "Sampled projections cluster around the sample point")
testutils.confirm_equal(worst_direction_stdev_observed,
worst_direction_stdev_predicted,
eps = 0.1,
worstcase = True,
relative = True,
msg = f"Predicted worst-case projections match sampled observations")
# I now compare the variances. The cross terms have lots of apparent error,
# but it's more meaningful to compare the eigenvectors and eigenvalues, so I
# just do that
testutils.confirm_covariances_equal(Var_q,
Var_q_observed,
what = "Var_q",
eps_eigenvalues = 0.05,
eps_eigenvectors_deg = 10,
check_biggest_eigenvalue_only = True)
if args.make_documentation_plots is not None:
def makeplot(**plotoptions):
gp.plot(*mrcal.utils._plot_args_points_and_covariance_ellipse( \
q_query_sampled,
"Observed in simulation"),
mrcal.utils._plot_arg_covariance_ellipse( \
np.mean(q_query_sampled, axis=0),
Var_q,
"Predicted by mrcal"),
square = 1,
xlabel = "x (pixels)",
ylabel = "y (pixels)",
title = "Surveyed calibration: distribution of point projections due to calibration-time noise",
**plotoptions)
if args.make_documentation_plots:
for extension in ('pdf','svg','png','gp'):
makeplot(wait = False,
terminal = shorter_terminal(terminal[extension]),
_set = extraset[extension],
hardcopy = f'{args.make_documentation_plots}/var-q.{extension}')
else:
makeplot(wait = True)
testutils.finish()