-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathmetrics.py
More file actions
169 lines (134 loc) · 5.38 KB
/
metrics.py
File metadata and controls
169 lines (134 loc) · 5.38 KB
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
import numpy as np
from collections import defaultdict
class ClasswiseAccuracy(object):
def __init__(self, num_classes):
self.num_classes = num_classes
self.tp_per_class = defaultdict(int)
self.count_per_class = defaultdict(int)
self.count = 0
self.tp = 0
def add_batch(self, y, y_hat):
for true, pred in zip(y, y_hat):
self.count_per_class["class_" + str(true.item())] += 1
self.count += 1
if true == pred:
self.tp_per_class["class_" + str(true.item())] += 1
self.tp += 1
def get_classwise_accuracy(self):
return {k: self.tp_per_class[k] / count for k, count in self.count_per_class.items()}
def get_average_accuracy(self):
cw_acc = self.get_classwise_accuracy()
return np.mean(list(cw_acc.values()))
def get_overall_accuracy(self):
return self.tp / self.count
class ClasswiseMultilabelMetrics(object):
def __init__(self, num_classes, prefix="class_"):
self.num_classes = num_classes
self.prefix = prefix
self.data = {prefix + str(i): {"tp": 0, "tn": 0, "fp": 0, "fn": 0} for i in range(num_classes)} # holds the classwise tp,tn,fp,fn
# hold the overall tp,tn,fp,fn
self.num_tp = 0
self.num_fp = 0
self.num_tn = 0
self.num_fn = 0
def add_batch(self, y_batch, y_hat_batch):
for y, y_hat in zip(y_batch, y_hat_batch):
for i in range(self.num_classes):
class_data = self.data[self.prefix + str(i)]
if y_hat[i] == y[i]:
# true-x
if y_hat[i]:
class_data["tp"] += 1
self.num_tp += 1
else:
class_data["tn"] += 1
self.num_tn += 1
else:
# false-x
if y_hat[i]:
class_data["fp"] += 1
self.num_fp += 1
else:
class_data["fn"] += 1
self.num_fn += 1
def get_classwise_precision(self):
# tp / (tp + fp)
out = {}
for i in range(self.num_classes):
class_data = self.data[self.prefix + str(i)]
if class_data["tp"] == 0:
out[self.prefix + str(i)] = 0
else:
out[self.prefix + str(i)] = class_data["tp"] / (class_data["tp"] + class_data["fp"])
return out
def get_classwise_recall(self):
# tp / (tp + fn)
out = {}
for i in range(self.num_classes):
class_data = self.data[self.prefix + str(i)]
if class_data["tp"] == 0:
out[self.prefix + str(i)] = 0
else:
out[self.prefix + str(i)] = class_data["tp"] / (class_data["tp"] + class_data["fn"])
return out
def get_classwise_f1(self):
# 2 * (precision * recall) / (precision + recall)
cw_prec = self.get_classwise_precision()
cw_rec = self.get_classwise_recall()
out = {}
for i in range(self.num_classes):
precision = cw_prec[self.prefix + str(i)]
recall = cw_rec[self.prefix + str(i)]
if (precision + recall) == 0:
out[self.prefix + str(i)] = 0
else:
out[self.prefix + str(i)] = 2 * (precision * recall) / (precision + recall)
return out
def get_average_f1(self):
cw_f1 = self.get_classwise_f1()
return np.mean(list(cw_f1.values()))
def get_average_recall(self):
cw_recall = self.get_classwise_recall()
return np.mean(list(cw_recall.values()))
def get_average_precision(self):
cw_precision = self.get_classwise_precision()
return np.mean(list(cw_precision.values()))
def get_overall_precision(self):
# tp / (tp + fp)
if self.num_tp == 0:
return 0
else:
return self.num_tp / (self.num_tp + self.num_fp)
def get_overall_recall(self):
# tp / (tp + fn)
if self.num_tp == 0:
return 0
else:
return self.num_tp / (self.num_tp + self.num_fn)
def get_overall_f1(self):
# 2 * (precision * recall) / (precision + recall)
precision = self.get_overall_precision()
recall = self.get_overall_recall()
if (precision + recall) == 0:
return 0
else:
return 2 * (precision * recall) / (precision + recall)
class PixelwiseMetrics(object):
def __init__(self, num_classes):
self.num_classes = num_classes
self.count = 0
self.data = {"pixelclass_" + str(i): {"acc": 0} for i in range(num_classes)}
def add_batch(self, y, y_hat):
self.count += 1
for c in range(self.num_classes):
class_data = self.data["pixelclass_" + str(c)]
preds_c = y_hat == c
targs_c = y == c
num_correct = (preds_c * targs_c).sum().cpu().detach().numpy()
num_pixels = np.sum(targs_c.cpu().detach().numpy())
class_data["acc"] += num_correct / num_pixels
def get_classwise_accuracy(self):
return {k: el['acc'] / self.count for k, el in self.data.items()}
def get_average_accuracy(self):
cw_acc = self.get_classwise_accuracy()
return np.mean(list(cw_acc.values()))