-
Notifications
You must be signed in to change notification settings - Fork 9
/
evaluation.py
189 lines (164 loc) · 7.97 KB
/
evaluation.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
import numpy as np
import model_util.pspnet.utils as utils
class Evaluator(object):
def __init__(self, dataset, is_training):
super(Evaluator, self).__init__()
self.is_training = is_training
self.dataset = dataset
def evaluate(self, prediction, ground_truth):
pass
def visualization(self, pri_v, prd_v, gt_v):
pass
def empty(self):
pass
class IoU_Evaluator(Evaluator):
def __init__(self, num_class, dataset, is_training, max_buffer_size=100000):
super(IoU_Evaluator, self).__init__(dataset, is_training)
self.num_class = num_class
self.threshold_dict = {0: 0.99,
1: 0.95,
2: 0.90,
3: 0.85,
4: 0.80,
5: 0.75,
6: 0.70,
7: 0.60,
8: 0.50,
9: 0.40,
10: 0.30
}
self._IoU_dict = {}
self._iIoU_dict = {}
self._IoU_acc = {}
self._iIoU_acc = {}
self._IoU_buffer = {}
self._iIoU_buffer = {}
self._max_buffer_size = max_buffer_size
self._count = {}
def __update(self):
if self.is_training:
for key in self._IoU_dict.keys():
if self._IoU_dict[key] == -1:
pass
else:
if key not in self._IoU_buffer:
self._IoU_buffer[key] = [self._IoU_dict[key]]
else:
if len(self._IoU_buffer[key]) < self._max_buffer_size:
self._IoU_buffer[key].append(self._IoU_dict[key])
else:
self._IoU_buffer[key].pop(0)
self._IoU_buffer[key].append(self._IoU_dict[key])
else:
for key in self._IoU_dict.keys():
if self._IoU_dict[key] == -1:
pass
else:
self._count[key] = \
self._count[key] + 1 if key in self._count else 0
self._IoU_acc[key] = \
self._IoU_acc[key] + self._IoU_dict[key] if key in self._IoU_acc else self._IoU_dict[key]
self._IoU_dict = {}
def __evaluate_once(self, class_prd_nuc, ground_truth):
unignored_region = ground_truth != 255
for i in range(self.num_class):
this_predicted_positive_region = class_prd_nuc == i
this_groudtruth_positive_region = ground_truth == i
TP_region = np.logical_and(this_predicted_positive_region, this_groudtruth_positive_region)
TPFPFN_region = np.logical_or(this_predicted_positive_region, this_groudtruth_positive_region)
TP = np.sum(np.logical_and(TP_region, unignored_region))
TPFPFN = np.sum(np.logical_and(TPFPFN_region, unignored_region))
self._IoU_dict[utils.ade20k_id2label[i].name] = TP / TPFPFN if TPFPFN > 0 else -1
def evaluate(self, prediction, ground_truth):
assert prediction.shape[-1] - self.num_class == 0, "number of predicted class mismatch"
def __preprocess():
class_prd = np.argmax(prediction, axis=-1)
# nuc -> none uncertainty
class_prd_nuc = np.copy(class_prd)
class_prd_nuc[class_prd_nuc == -1] = \
np.argsort(-prediction, axis=-1)[:, :, 1][class_prd_nuc == -1]
return class_prd_nuc
class_prd_nuc = __preprocess()
self.__evaluate_once(class_prd_nuc, ground_truth)
self.__update()
def empty(self):
self._IoU_buffer = {}
def visualization(self, prd_v, NSactitp_v, gt_v, pri_v=None):
threshold_dict = {0: 0.99,
1: 0.95,
2: 0.90,
3: 0.85,
4: 0.80,
5: 0.75,
6: 0.70,
7: 0.60,
8: 0.50,
9: 0.40,
10: 0.30
}
def __compute_colored_gt_image():
colored_gt_image = utils.color_class_image(gt_v, self.dataset)
colored_gt_image = np.expand_dims(colored_gt_image, axis=0)
return colored_gt_image
def __compute_colored_pri_image():
class_image_prior = np.argmax(pri_v, axis=-1)
class_image_prior[gt_v == 255] = 255
class_image_prior_neq = np.copy(class_image_prior)
class_image_prior_neq[class_image_prior_neq != gt_v] = -1
probability_image_prior = np.max(pri_v, axis=-1)
class_image_prior_threshold = [class_image_prior, class_image_prior_neq]
for i in range(0, 11):
class_image_prior_copy = np.copy(class_image_prior)
class_image_prior_copy[probability_image_prior < threshold_dict[i]] = -1
class_image_prior_copy[gt_v == 255] = 255
class_image_prior_threshold.append(class_image_prior_copy)
colored_pri_images = []
for i in range(0, 13):
this_class_image_prior = class_image_prior_threshold[i]
colored_pri_images.append(
utils.color_class_image(this_class_image_prior, self.dataset))
colored_pri_images = np.asarray(colored_pri_images, dtype=np.uint8)
return colored_pri_images
def __compute_colored_prd_image():
class_image_prd_nuc = np.argmax(prd_v, axis=-1)
actprd =np.squeeze(
NSactitp_v * (np.expand_dims(class_image_prd_nuc, axis=-1) + 1) - 1,
axis=-1
)
colored_prd_images = []
for i in range(0, actprd.shape[0]):
this_class_image_prd_nuc = class_image_prd_nuc[i]
this_class_image_prd_nuc[gt_v == 255] = 255
colored_prd_images.append(
utils.color_class_image(this_class_image_prd_nuc, self.dataset))
this_class_image_prd_nuc_neq = np.copy(this_class_image_prd_nuc)
this_class_image_prd_nuc_neq[this_class_image_prd_nuc_neq != gt_v] = -1
this_class_image_prd_nuc_neq[gt_v == 255] = 255
colored_prd_images.append(
utils.color_class_image(this_class_image_prd_nuc_neq, self.dataset))
this_class_image_actprd = actprd[i]
this_class_image_actprd[gt_v == 255] = 255
colored_prd_images.append(
utils.color_class_image(this_class_image_actprd, self.dataset))
colored_prd_images = np.asarray(colored_prd_images, dtype=np.uint8)
return colored_prd_images
colored_gt_image = __compute_colored_gt_image()
colored_prd_images = __compute_colored_prd_image()
if pri_v is not None:
colored_pri_images = __compute_colored_pri_image()
return colored_pri_images, colored_prd_images, colored_gt_image
else:
return colored_prd_images, colored_gt_image
@property
def IoU(self):
IoU_mean = {'class_mean':[]}
if self.is_training:
for key in self._IoU_buffer.keys():
IoU_mean[key] = sum(self._IoU_buffer[key]) / len(self._IoU_buffer[key])
IoU_mean['class_mean'] += self._IoU_buffer[key]
else:
for key in self._IoU_buffer.keys():
IoU_mean[key] = self._IoU_acc[key] / self._count[key]
IoU_mean['class_mean'] += self._IoU_buffer[key]
IoU_mean['class_mean'] = sum(IoU_mean['class_mean']) / len(IoU_mean['class_mean'])
return IoU_mean