-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluate.py
521 lines (399 loc) · 21.7 KB
/
evaluate.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
from util.options import *
from util.forward import *
from util.fitting import *
from util.initialisation import *
from util import consistency
from datasets.nyu_depth.nyu_depth import NYURGBDataset
import numpy as np
import torch
import matplotlib.pyplot as plt
from matplotlib import rc
rc('text', usetex=False)
plt.rc("font", size=8, family="serif")
import random
import skimage.transform
import time
import platform
import sklearn.metrics
import pickle
total_start = time.time()
opt = get_options()
opt = load_opts_for_eval(opt)
opt.batch = 1
opt.samplecount = 1
if opt.seed < 0:
opt.seed = int(np.random.uniform(0, 100000))
eval_folder, log = create_eval_folder(opt)
max_auc_distances = [200, 100, 50, 20, 10, 5]
auc_values = {}
best_auc_values = {}
best_auc_value_idx = {}
for dist in max_auc_distances:
auc_values["auc_at_%d" % dist] = []
best_auc_values["auc_at_%d" % dist] = 0
best_auc_value_idx["auc_at_%d" % dist] = None
auc_values["mean"] = []
auc_values["mean_chamfer"] = []
hostname = platform.node()
print("host: ", hostname)
print("SLURM job ID: ", opt.jobid)
torch.manual_seed(opt.seed)
np.random.seed(opt.seed)
random.seed(opt.seed)
if opt.dataset == 'nyu':
valset = NYURGBDataset(data_directory=opt.data_path, split=opt.split, scale=1, split_mat=opt.nyu_split)
depth_mean = valset.depth_mean
depth_stdd = valset.depth_stdd
coord_mean = torch.Tensor([-0.03198672, -0.20830469, 2.7163548])
image_mean = torch.from_numpy(valset.image_mean)
else:
assert False, "unknown dataset %s" % opt.dataset
valset_loader = torch.utils.data.DataLoader(valset, shuffle=False, num_workers=6, batch_size=1)
print_options(opt)
devices = get_devices(opt)
fitting_device, consac_device, depth_device, inlier_device = devices
depth_model = get_depth_model(opt, devices)
feed_height = depth_model["height"]
feed_width = depth_model["width"]
inlier_fun = consistency.soft_inlier_fun_gen(5. / opt.threshold, opt.threshold)
if opt.lbfgs:
minimal_solver = CuboidFitLBFGS(a_max=opt.a_max, norm_by_volume=True)
else:
minimal_solver = CuboidFitAdam(a_max=opt.a_max, norm_by_volume=True)
all_losses = []
image_idx = 0
all_miss_rates = []
miss_rates_per_image = []
H, W, Y, H_, W_, Y_, M, P, S, Q, R, B, K, model_dim, data_dim, minimal_set_size, dimensions = get_dimensions(opt, valset)
consac = get_consac_model(opt, devices)
consistency_fun = consistency.cuboids_consistency_measure_iofun_parallel_batched_torch
torch.set_grad_enabled(False)
all_chamfer_distances = []
lists_of_all_distances = [[] for ki in range(K)]
lists_of_chamfer_distances = [[] for ki in range(K)]
lists_of_oa_distances = [[] for ki in range(K)]
list_of_mean_weighted_mses = []
print("\n##########\n")
for (image, intrinsic, true_coord_grid, labels, gt_models, gt_depth, file_indices) in valset_loader:
if image_idx < opt.sampleid:
print("%d.. " % image_idx, end="\r")
image_idx += 1
continue
print("image index: %d / %d " % (image_idx, len(valset_loader)))
sample_results = {}
bi = 0
depth, depth_normalised, depth_mse = \
estimate_depth(opt, image, image_mean, depth_model, dimensions, devices, depth_mean, depth_stdd, gt_depth,
read_cache=opt.read_cache, write_cache=opt.write_cache)
if opt.write_cache_only:
image_idx += 1
continue
data = depth_normalised.detach().to(consac_device)
true_coord_grid, true_coord_grid_small, true_coord_flat, \
estm_coord_grid, estm_coord_grid_small, estm_coord_flat = \
generate_coordinate_grids(depth, true_coord_grid, dimensions, devices)
single_start = time.time()
states = torch.zeros((P, K, B, H, W, 1), device=consac_device)
all_best_inliers_estm = [torch.zeros((P, K, B, Y_), device=fitting_device) for _ in range(M)]
all_best_inliers_gt = [torch.zeros((P, K, B, Y_), device=depth_device) for _ in range(M)]
all_probs = torch.zeros((P, M, K, B, Q, R, H_, W_), device=consac_device)
all_q_probs = torch.zeros((P, M, K, B, Q), device=consac_device)
all_best_models = torch.zeros((P, M, K, B, model_dim), device=fitting_device)
all_best_inlier_counts_estm = torch.zeros((P, M, K, B), device=depth_device)
all_inlier_counts_estm = torch.zeros((P, M, S, K, B), device=depth_device)
all_inlier_counts_gt = torch.zeros((P, M, S, K, B), device=depth_device)
mean_oa_distances_gt = torch.zeros((P, M, S, K, B), device=depth_device)
all_best_inliers_estm_tensor = torch.zeros((P, M, K, B, Y_), device=fitting_device)
all_best_inliers_gt_tensor = torch.zeros((P, M, K, B, Y_), device=fitting_device)
max_distance_to_occlusions_gt = None
max_distance_to_occlusions_estm = None
prev_inliers_estm = None
prev_inliers_gt = None
prev_distances_gt = None
prev_occluded_distances_gt = None
for mi in range(M):
print("fitting cuboid %d" % mi, end="\r")
if prev_inliers_estm is not None:
for pi in range(P):
for ki in range(K):
inliers_scaled = torch.nn.functional.interpolate(
prev_inliers_estm[pi, ki, :].view(B, 1, H_, W_), size=(H, W)).squeeze()
states[pi, :, :, :, :, 0] = inliers_scaled
sampling_weight_maps, selection_weights, log_probs, log_q, entropy = \
estimate_sampling_weights(opt, dimensions, devices, data, states, consac)
all_probs[:, mi, :] = sampling_weight_maps.view(P, S, K, B, Q, R, H_, W_)[:, 0]
all_q_probs[:, mi, :] = selection_weights[:, 0]
models, choices, sel_choices, residual = \
estimate_models(opt, dimensions, estm_coord_flat, sampling_weight_maps[:, :, :, :, :Q].detach(),
selection_weights.detach(), minimal_solver)
inliers_estm, distances_estm, occluded_distances_estm = \
count_inliers(opt, models, estm_coord_flat, inlier_fun, None, None, prev_inliers_estm,
occlusion_aware=(not opt.no_oai_sampling))
del distances_estm, occluded_distances_estm
inliers_gt, distances_gt, occluded_distances_gt = \
count_inliers(opt, models, true_coord_flat, inlier_fun, prev_distances_gt,
prev_occluded_distances_gt, prev_inliers_gt, occlusion_aware=(not opt.no_oai_loss))
best_single_hypos, best_inliers_estm, all_best_models[:, mi] = \
select_single_hypotheses(opt, dimensions, inliers_estm, models=models)
prev_inliers_estm = torch.gather(inliers_estm, 1, best_single_hypos.view(P, 1, K, B, 1).expand(P, 1, K, B, Y_))
all_best_inliers_estm_tensor[:, mi] = prev_inliers_estm
prev_inliers_gt = torch.gather(inliers_gt, 1, best_single_hypos.view(P, 1, K, B, 1).expand(P, 1, K, B, Y_))
prev_distances_gt = torch.gather(distances_gt, 1, best_single_hypos.view(P, 1, K, B, 1).expand(P, 1, K, B, Y_))
prev_occluded_distances_gt = torch.gather(occluded_distances_gt, 1,
best_single_hypos.view(P, 1, K, B, 1).expand(P, 1, K, B, Y_))
oa_distances_gt = torch.max(distances_gt, occluded_distances_gt)
all_best_inliers_gt_tensor[:, mi] = prev_inliers_gt
all_inlier_counts_estm[:, mi] = inliers_estm.sum(-1).to(all_inlier_counts_estm.device)
all_inlier_counts_gt[:, mi] = inliers_gt.sum(-1).to(all_inlier_counts_gt.device)
mean_oa_distances_gt[:, mi] = oa_distances_gt.to(mean_oa_distances_gt.device).mean(dim=-1)
all_best_oa_distances_gt = \
torch.gather(oa_distances_gt, 1,
best_single_hypos.view(P, 1, K, B, 1).expand(P, 1, K, B, Y_).to(
oa_distances_gt.device)).squeeze(1)
all_best_inlier_counts_estm[:, mi] = \
torch.gather(all_inlier_counts_estm[:, mi], 1,
best_single_hypos.view(P, 1, K, B).to(all_inlier_counts_estm.device)).squeeze(1)
all_num_primitives = torch.ones((P, K, B), device=all_best_inlier_counts_estm.device, dtype=torch.long)
selected_joint_inlier_counts = all_best_inlier_counts_estm[:, 0]
for bi in range(B):
for pi in range(P):
for ki in range(K):
for mi in range(1, M):
inlier_increase = all_best_inlier_counts_estm[pi, mi, ki, bi] - \
all_best_inlier_counts_estm[pi, mi-1, ki, bi]
print("cuboid %d : inlier increase: " % mi, inlier_increase.item())
if inlier_increase.item() < opt.inlier_cutoff:
break
else:
selected_joint_inlier_counts[pi, ki, bi] = all_best_inlier_counts_estm[pi, mi, ki, bi]
all_num_primitives[pi, ki, bi] += 1
best_outer_hypos = torch.argmax(selected_joint_inlier_counts, dim=0) # KxB
best_num_primitives = torch.gather(all_num_primitives, 0, best_outer_hypos.view(1, K, B).expand(1, K, B)).squeeze(0)
best_models = torch.gather(all_best_models.to(best_outer_hypos.device), 0, best_outer_hypos.view(1, 1, K, B, 1).expand(1, M, K, B, model_dim)).squeeze(0)
distances_gt = \
consistency.cuboids_consistency_measure_iofun_parallel_batched_torch(
best_models.to(true_coord_grid.device).unsqueeze(0), true_coord_grid.view(B, Y, 3)).squeeze(0)
distances_gt = torch.sqrt(distances_gt)
min_distances_gt, _ = torch.min(distances_gt, dim=0)
for ki in range(K):
for bi in range(B):
min_distances_gt[ki, bi], _ = torch.min(distances_gt[:best_num_primitives[ki, bi], ki, bi], dim=0)
chamfer_distances_gt = torch.mean(min_distances_gt, dim=-1) # KxB
all_chamfer_distances += [chamfer_distances_gt.detach().cpu().numpy()]
models_batched = best_models.view(-1, model_dim).detach().to(inlier_device)
features_expanded = true_coord_grid.view(B, Y, 3).view(1, 1, B, Y, 3).expand(M, K, B, Y, 3).to(inlier_device)
features_batched = features_expanded.contiguous().view(-1, Y, 3).detach()
occlusions_gt_batched, closest_sides_gt_batched, distances_to_cube_sides_gt_batched, _ = \
consistency.cuboid_occlusion_torch(models_batched, features_batched)
occlusions_gt = occlusions_gt_batched.view(M, K, B, Y, 6)
distances_to_sides_gt = distances_to_cube_sides_gt_batched.view(M, K, B, Y, 6)
occlusion_distances_gt = torch.sqrt(torch.max(torch.max(occlusions_gt * distances_to_sides_gt, dim=-1)[0], dim=0)[0])
depth_error = (depth[bi, :, :].cpu().detach().numpy().squeeze() - gt_depth[bi].cpu().detach().numpy().squeeze()) ** 2
depth_mse = np.mean(depth_error)
sample_results["image"] = image.cpu().detach().numpy()
sample_results["depth"] = depth.cpu().detach().numpy()
sample_results["depth_normalised"] = depth_normalised.cpu().detach().numpy()
sample_results["depth_mse"] = depth_mse
sample_results["all_best_models"] = all_best_models.cpu().detach().numpy()
sample_results["all_num_primitives"] = all_num_primitives.cpu().detach().numpy()
sample_results["best_models"] = best_models.cpu().detach().numpy()
sample_results["best_num_primitives"] = best_num_primitives.cpu().detach().numpy()
sample_results["best_outer_hypos"] = best_outer_hypos.cpu().detach().numpy()
if opt.save_all:
sample_results["probability_maps"] = all_probs.cpu().detach().numpy()
sample_results["selection_probabilities"] = all_q_probs.cpu().detach().numpy()
sample_results["distances_to_sides_gt"] = distances_to_sides_gt.cpu().detach().numpy()
sample_results["occlusions_gt"] = occlusions_gt.cpu().detach().numpy()
sample_results["chamfer_distances_gt"] = chamfer_distances_gt.cpu().detach().numpy()
sample_results["occlusion_distances_gt"] = occlusion_distances_gt.cpu().detach().numpy()
sample_results["distances_gt"] = distances_gt.cpu().detach().numpy()
sample_results["min_distances_gt"] = min_distances_gt.cpu().detach().numpy()
for ki in range(K):
lists_of_chamfer_distances[ki] += [chamfer_distances_gt.detach().cpu().numpy()[ki, 0]]
min_distances_gt_numpy = min_distances_gt[ki].cpu().detach().numpy().squeeze()
mean_chamfer = np.mean(min_distances_gt_numpy)
occlusion_distances_gt_numpy = occlusion_distances_gt[ki].cpu().detach().numpy().squeeze()
min_distances_gt_numpy = np.maximum(min_distances_gt_numpy, occlusion_distances_gt_numpy)
mean_distance = np.mean(min_distances_gt_numpy)
print("mean OA distance: %.3f" % mean_distance)
lists_of_oa_distances[ki] += [mean_distance]
all_distances_sorted = np.sort(min_distances_gt_numpy)
lists_of_all_distances[ki] += all_distances_sorted.tolist()
inlier_range = np.arange(Y).astype(np.float32) * 1. / Y
sample_auc_values = {}
for max_distance in max_auc_distances:
max_value = max_distance / 100.
x = np.append(all_distances_sorted[np.where(all_distances_sorted < max_value)], max_value)
y = inlier_range[np.where(all_distances_sorted < max_value)]
if y.size > 2:
y = np.append(y, y[-1])
auc = sklearn.metrics.auc(x, y) / max_value
else:
auc = 0
sample_auc_values["auc_at_%d" % max_distance] = auc
auc_values["auc_at_%d" % max_distance] += [auc]
if auc > best_auc_values["auc_at_%d" % max_distance]:
best_auc_values["auc_at_%d" % max_distance] = auc
best_auc_value_idx["auc_at_%d" % max_distance] = {"ki": ki, "idx": image_idx, "fidx": file_indices[0].item()}
sample_auc_values["mean"] = mean_distance
sample_auc_values["mean_chamfer"] = mean_chamfer
auc_values["mean"] += [mean_distance]
auc_values["mean_chamfer"] += [mean_chamfer]
sample_results["sample_%d" % ki] = {"chamfer": chamfer_distances_gt[ki].squeeze().detach().cpu().numpy(),
"auc": sample_auc_values, "inlier_range": inlier_range,
"all_distances_sorted": all_distances_sorted}
if eval_folder is not None:
target_file = os.path.join(eval_folder, "sample_%04d_%04d.pkl" % (image_idx, file_indices[0].item()))
pickle.dump(sample_results, open(target_file, "wb"))
single_time = time.time() - single_start
print("time elapsed: %.2f s" % single_time)
depth_errors = gt_depth[bi].cpu().detach().numpy().squeeze() - depth[bi, :, :].cpu().detach().numpy().squeeze()
depth_errors *= depth_errors
all_probs_np = all_probs[best_outer_hypos].cpu().detach().numpy().squeeze()
all_q_probs_np = all_q_probs[best_outer_hypos].cpu().detach().numpy().squeeze()
all_weighted_mses = []
gray_image = rgb2gray(image[bi].cpu().detach().numpy().squeeze())
gray_im_scaled = (gray_image - 0.5) * 0.25
for mi in range(M):
probs = all_probs_np[mi, :]
q_probs = all_q_probs_np[mi, :]
weighted_mses = []
for qi in range(Q):
probs_ = probs[qi]
probs_ = probs_ / np.max(probs_)
probs_ = skimage.transform.resize(probs_, gray_im_scaled.shape)
weighted_depth_errors = depth_errors * (probs_)
weighted_mse = np.sum(depth_errors) / np.sum(probs_)
q = q_probs[qi]
weighted_mses += [weighted_mse * q]
all_weighted_mses += [np.sum(weighted_mses)]
mean_weighted_mse = np.mean(all_weighted_mses)
list_of_mean_weighted_mses += [mean_weighted_mse]
if opt.visualise:
all_probs_np = all_probs[best_outer_hypos].cpu().detach().numpy().squeeze()
all_q_probs_np = all_q_probs[best_outer_hypos].cpu().detach().numpy().squeeze()
all_best_inliers_estm_np = all_best_inliers_estm_tensor[best_outer_hypos].view(M, H_, W_).cpu().detach().numpy()
all_best_inliers_gt_np = all_best_inliers_gt_tensor[best_outer_hypos].view(M, H_, W_).cpu().detach().numpy()
all_best_inliers_estm_torch = all_best_inliers_estm_tensor[best_outer_hypos].view(M, 1, H_, W_)
all_best_inliers_estm_torch = torch.nn.functional.interpolate(all_best_inliers_estm_torch,
size=(H, W)).squeeze()
all_best_inliers_gt_torch = all_best_inliers_gt_tensor[best_outer_hypos].view(M, 1, H_, W_)
all_best_inliers_gt_torch = torch.nn.functional.interpolate(all_best_inliers_gt_torch, size=(H, W)).squeeze()
colours = ['#000000', '#e6194b', '#4363d8', '#aaffc3', '#911eb4', '#46f0f0', '#f58231', '#3cb44b', '#f032e6',
'#008080', '#bcf60c', '#fabebe', '#e6beff', '#9a6324', '#fffac8', '#800000', '#aaffc3']
points = depth[bi].cpu().numpy()
fig = plt.figure()
plt.axis('off')
# plt.tight_layout(pad=0.2)
subplot_id = 1
Vsp = np.maximum(4, Q+2)
Hsp = M+1
ax0a = fig.add_subplot(Vsp, M+1, 1)
ax0a.axis('off')
ax0a.imshow(depth[bi, :, :].cpu().detach().numpy().squeeze())
ax0b = fig.add_subplot(Vsp, M+1, M+2)
ax0b.imshow(depth_error)
ax0b.set_title("MSE: %.3f" % depth_mse)
ax0b.axis('off')
ax0c = fig.add_subplot(Vsp, Hsp, 2*(M+1)+1)
ax0c.axis('off')
ax0c.imshow(gt_depth[bi].cpu().detach().numpy().squeeze())
ax0d = fig.add_subplot(Vsp, Hsp, 3*(M+1)+1)
ax0d.axis('off')
ax0d.imshow(image[bi].cpu().detach().numpy().squeeze())
gray_image = rgb2gray(image[bi].cpu().detach().numpy().squeeze())
gray_im_scaled = (gray_image - 0.5) * 0.25
for mi in range(M):
points = estm_coord_flat
probs = all_probs_np[mi, :]
q_probs = all_q_probs_np[mi, :]
inliers_estm = all_best_inliers_estm_torch[mi]
inliers_gt = all_best_inliers_gt_torch[mi]
for qi in range(Q):
ax1 = fig.add_subplot(Vsp, Hsp, qi*(M+1) + mi + 2)
ax1.axis('off')
probs_ = probs[qi]
probs_ = probs_ / np.max(probs_)
probs_ = skimage.transform.resize(probs_, gray_im_scaled.shape)
weighted_depth_errors = depth_errors * (probs_)
weighted_mse = np.sum(depth_errors) / np.sum(probs_)
q = q_probs[qi]
ax1.imshow(probs_ + gray_im_scaled)
ax1.set_title("%.3f" % q)
ax2 = fig.add_subplot(Vsp, Hsp, Q*(Hsp) + mi + 2)
ax3 = fig.add_subplot(Vsp, Hsp, (Q+1)*Hsp + mi + 2)
ax2.axis('off')
ax3.axis('off')
colour_list = []
inliers_estm = inliers_estm.detach().cpu().numpy()
inliers_gt = inliers_gt.detach().cpu().numpy()
ax2.imshow(inliers_estm + gray_im_scaled, vmin=-1.125, vmax=1.125)
ax2.set_title("estimated inliers")
ax3.imshow(inliers_gt + gray_im_scaled, vmin=-1.125, vmax=1.125)
ax3.set_title("true inliers")
plt.show()
image_idx += 1
print("\nMean L2-Distances:")
chamfer_per_run = []
for ki in range(K):
chamfer_mean = np.mean(lists_of_chamfer_distances[ki])
chamfer_stdd = np.std(lists_of_chamfer_distances[ki])
print("- run %d : %.6f (%.6f)" % (ki, chamfer_mean, chamfer_stdd))
chamfer_per_run += [chamfer_mean]
print("-- per run mean: %.6f" % np.mean(chamfer_per_run))
print("-- per run stdd: %.6f" % np.std(chamfer_per_run))
print("\nMean Occlusion-Aware Distances:")
chamfer_per_run = []
for ki in range(K):
chamfer_mean = np.mean(lists_of_oa_distances[ki])
chamfer_stdd = np.std(lists_of_oa_distances[ki])
print("- run %d : %.6f (%.6f)" % (ki, chamfer_mean, chamfer_stdd))
chamfer_per_run += [chamfer_mean]
print("-- per run mean: %.6f" % np.mean(chamfer_per_run))
print("-- per run stdd: %.6f" % np.std(chamfer_per_run))
list_of_all_distances = []
for ki in range(len(lists_of_all_distances)):
list_of_all_distances += lists_of_all_distances[ki]
print("\nAUC for Occlusion-Aware Distances:")
for max_distance in max_auc_distances:
max_value = max_distance / 100.
print("- AUC at %d" % max_distance)
auc_per_run_list = []
for ki in range(K):
num_distances = len(lists_of_all_distances[ki])
all_distances_sorted = np.sort(np.array(lists_of_all_distances[ki]))
inlier_range = np.arange(num_distances).astype(np.float32) * 1. / num_distances
x = np.append(all_distances_sorted[np.where(all_distances_sorted < max_value)], max_value)
y = inlier_range[np.where(all_distances_sorted < max_value)]
if y.size > 2:
y = np.append(y, y[-1])
auc = sklearn.metrics.auc(x, y) / max_value
else:
auc = 0
print("-- run %d : %.6f" % (ki, auc))
auc_per_run_list += [auc]
print("--- mean: %.6f" % np.mean(auc_per_run_list))
print("--- stdd: %.6f" % np.std(auc_per_run_list))
num_distances = len(list_of_all_distances)
all_distances_sorted = np.sort(np.array(list_of_all_distances))
inlier_range = np.arange(num_distances).astype(np.float32) * 1. / num_distances
x = np.append(all_distances_sorted[np.where(all_distances_sorted < max_value)], max_value)
y = inlier_range[np.where(all_distances_sorted < max_value)]
if y.size > 2:
y = np.append(y, y[-1])
auc = sklearn.metrics.auc(x, y) / max_value
else:
auc = 0
print("---- overall AUC: %.6f" % auc)
total_time = time.time() - total_start
print("\ntime elapsed: %.1f s (%.2f s)" % (total_time, total_time*1./len(valset_loader)))
all_results = {}
all_results["auc_values"] = auc_values
all_results["best_auc_values"] = best_auc_values
all_results["best_auc_value_idx"] = best_auc_value_idx
all_results["all_chamfer_distances"] = all_chamfer_distances
all_results["lists_of_all_distances"] = lists_of_all_distances
all_results["lists_of_chamfer_distances"] = lists_of_chamfer_distances
if eval_folder is not None:
target_file = os.path.join(eval_folder, "results.pkl")
pickle.dump(all_results, open(target_file, "wb"))