-
Notifications
You must be signed in to change notification settings - Fork 7
/
hybrid_evaluate_depth.py
233 lines (192 loc) · 7.97 KB
/
hybrid_evaluate_depth.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
from __future__ import absolute_import, division, print_function
from open3d import *
import os
import cv2
import numpy as np
import torch
from utils import write_ply, backproject_depth, v, npy, Thres_metrics_np
cv2.setNumThreads(
0) # This speeds up evaluation 5x on our unix systems (OpenCV 3.3.1)
def compute_errors_perimage(gt, pred, min_depth, max_depth):
valid_mask = (gt > min_depth) & (gt < max_depth)
epe = np.mean(np.abs(gt[valid_mask] - pred[valid_mask]))
abs_rel = np.mean(np.abs(gt[valid_mask] - pred[valid_mask]) / gt[valid_mask])
sq_rel = np.mean(((gt[valid_mask] - pred[valid_mask])**2) / gt[valid_mask])
rmse = (gt - pred)**2
rmse = np.sqrt(rmse.mean())
rmse_log = (np.log(gt) - np.log(pred))**2
rmse_log = np.sqrt(rmse_log.mean())
thresh = np.maximum((gt / pred), (pred / gt))
a1 = (thresh < 1.25).mean()
a2 = (thresh < 1.25**2).mean()
a3 = (thresh < 1.25**3).mean()
log10 = np.mean(np.abs(np.log10(pred) - np.log10(gt)))
return {
'abs_rel':abs_rel.item(),
'sq_rel':sq_rel.item(),
'rmse':rmse.item(),
'rmse_log':rmse_log.item(),
'a1':a1.item(),
'a2':a2.item(),
'a3':a3.item(),
'log10':log10.item(),
'valid_number':1.0,
'abs_diff':epe.item()
}
def compute_errors(gt, pred, disable_median_scaling, min_depth, max_depth,
interval):
"""Computation of error metrics between predicted and ground truth depths
"""
# if not disable_median_scaling:
# ratio = np.median(gt) / np.median(pred)
# pred *= ratio
# pred[pred < min_depth] = min_depth
# pred[pred > max_depth] = max_depth
mask = np.logical_and(gt > min_depth, gt < max_depth)
thresh = np.maximum((gt / pred), (pred / gt))
a1 = (thresh < 1.25).mean()
a2 = (thresh < 1.25**2).mean()
a3 = (thresh < 1.25**3).mean()
rmse = (gt - pred)**2
rmse = np.sqrt(rmse.mean())
rmse_log = (np.log(gt) - np.log(pred))**2
rmse_log = np.sqrt(rmse_log.mean())
abs_rel = np.mean(np.abs(gt[mask] - pred[mask]) / gt[mask])
print('1',abs_rel)
abs_rel_2 = np.sum(np.abs(gt[mask] - pred[mask]) / gt[mask])/np.sum(mask.astype(np.float32))
print('2', abs_rel_2)
abs_diff = np.mean(np.abs(gt - pred))
# abs_diff_median = np.median(np.abs(gt - pred))
sq_rel = np.mean(((gt - pred)**2) / gt)
log10 = np.mean(np.abs(np.log10(pred) - np.log10(gt)))
# mask = np.ones_like(pred)
# thre1 = Thres_metrics_np(pred, gt, mask, 1.0, 0.2)
# thre3 = Thres_metrics_np(pred, gt, mask, 1.0, 0.5)
# thre5 = Thres_metrics_np(pred, gt, mask, 1.0, 1.0)
result = {}
result['abs_rel'] = abs_rel
result['sq_rel'] = sq_rel
result['rmse'] = rmse
result['rmse_log'] = rmse_log
result['log10'] = log10
result['a1'] = a1
result['a2'] = a2
result['a3'] = a3
result['abs_diff'] = abs_diff
result['total_count'] = 1.0
return result
def compute_errors1(gt, pred, disable_median_scaling, min_depth, max_depth,
interval):
"""Computation of error metrics between predicted and ground truth depths
"""
# if not disable_median_scaling:
# ratio = np.median(gt) / np.median(pred)
# pred *= ratio
# pred[pred < min_depth] = min_depth
# pred[pred > max_depth] = max_depth
thresh = np.maximum((gt / pred), (pred / gt))
a1 = (thresh < 1.25).mean()
a2 = (thresh < 1.25**2).mean()
a3 = (thresh < 1.25**3).mean()
rmse = (gt - pred)**2
rmse = np.sqrt(rmse.mean())
rmse_log = (np.log(gt) - np.log(pred))**2
rmse_log = np.sqrt(rmse_log.mean())
abs_rel = np.mean(np.abs(gt - pred) / gt)
abs_diff = np.mean(np.abs(gt - pred))
# abs_diff_median = np.median(np.abs(gt - pred))
sq_rel = np.mean(((gt - pred)**2) / gt)
log10 = np.mean(np.abs(np.log10(pred) - np.log10(gt)))
# mask = np.ones_like(pred)
# thre1 = Thres_metrics_np(pred, gt, mask, 1.0, 0.2)
# thre3 = Thres_metrics_np(pred, gt, mask, 1.0, 0.5)
# thre5 = Thres_metrics_np(pred, gt, mask, 1.0, 1.0)
result = {}
result['abs_rel'] = abs_rel
result['sq_rel'] = sq_rel
result['rmse'] = rmse
result['rmse_log'] = rmse_log
result['log10'] = log10
result['a1'] = a1
result['a2'] = a2
result['a3'] = a3
result['abs_diff'] = abs_diff
result['total_count'] = 1.0
return result
# return abs_rel, sq_rel, log10, rmse, rmse_log, a1, a2, a3, abs_diff, abs_diff_median, thre1, thre3, thre5
def evaluate_depth_maps(results, config, do_print=False):
errors = []
if not os.path.exists(config.save_dir):
os.makedirs(config.save_dir)
print('eval against gt depth map of size: %sx%d' %
(results[0][1].shape[0], results[0][1].shape[1]))
for i in range(len(results)):
if i % 100 == 0:
print('evaluation : %d/%d' % (i, len(results)))
gt_depth = results[i][1]
gt_height, gt_width = gt_depth.shape[:2]
pred_depth = results[i][0]
filename = results[i][2]
inv_K = results[i][3]
if gt_width != pred_depth.shape[1] or gt_height != pred_depth.shape[0]:
pred_depth = cv2.resize(pred_depth, (gt_width, gt_height),
interpolation=cv2.INTER_NEAREST)
mask = np.logical_and(gt_depth > config.MIN_DEPTH,
gt_depth < config.MAX_DEPTH)
if not mask.sum():
continue
ind = np.where(mask.flatten())[0]
if config.vis:
cam_points = backproject_depth(pred_depth, inv_K, mask=False)
cam_points_gt = backproject_depth(gt_depth, inv_K, mask=False)
write_ply('%s/%s_pred.ply' % (config.save_dir, filename),
cam_points[ind])
write_ply('%s/%s_gt.ply' % (config.save_dir, filename),
cam_points_gt[ind])
dataset = filename.split('_')[0]
interval = (935 - 425) / (128 - 1) # Interval value used by MVSNet
errors.append(
(compute_errors(gt_depth[mask], pred_depth[mask],
config.disable_median_scaling, config.MIN_DEPTH,
config.MAX_DEPTH, interval), dataset, filename))
with open('%s/errors.txt' % (config.save_dir), 'w') as f:
for x, _, fID in errors:
tex = fID + ' ' + ' '.join(['%.3f' % y for y in x])
f.write(tex + '\n')
np.save('%s/error.npy' % config.save_dir, errors)
results = {}
all_errors = [x[0] for x in errors]
print(f"total example evaluated: {len(all_errors)}")
all_mean_errors = np.array(all_errors).mean(0)
if do_print:
print("\n all")
print("\n " +
("{:>8} | " *
13).format("abs_rel", "sq_rel", "log10", "rmse", "rmse_log",
"a1", "a2", "a3", "abs_diff", "abs_diff_median"))
print(("&{: 8.3f} " * 13).format(*all_mean_errors.tolist()) + "\\\\")
error_names = [
"abs_rel", "sq_rel", "log10", "rmse", "rmse_log", "a1", "a2", "a3",
"abs_diff", "abs_diff_median", "thre1", "thre3", "thre5"
]
results['depth'] = {'error_names': error_names, 'errors': all_mean_errors}
errors_per_dataset = {}
for x in errors:
key = x[1]
if key not in errors_per_dataset:
errors_per_dataset[key] = [x[0]]
else:
errors_per_dataset[key].append(x[0])
if config.print_per_dataset_stats:
for key in errors_per_dataset.keys():
errors_ = errors_per_dataset[key]
mean_errors = np.array(errors_).mean(0)
print("\n dataset %s: %d" % (key, len(errors_)))
print("\n " +
("{:>8} | " *
13).format("abs_rel", "sq_rel", "log10", "rmse", "rmse_log",
"a1", "a2", "a3", "abs_diff", "abs_diff_median",
"thre1", "thre3", "thre5"))
print(("&{: 8.3f} " * 13).format(*mean_errors.tolist()) + "\\\\")
print("\n-> Done!")
return results