File size: 4,187 Bytes
a256709
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import numpy as np
import math


def f2_score(y_true, y_pred, threshold=0.5):
    return fbeta_score(y_true, y_pred, 2, threshold)


def dice(logit, truth, threshold=0.5):
    batch_size = len(truth)

    with torch.no_grad():
        logit = logit.view(batch_size, -1)
        truth = truth.view(batch_size, -1)
        assert logit.shape == truth.shape

        probability = logit
        p = (probability > threshold).float()
        t = (truth > 0.5).float()

        t_sum = t.sum(-1)
        p_sum = p.sum(-1)
        neg_index = torch.nonzero(t_sum == 0)
        pos_index = torch.nonzero(t_sum >= 1)
        # print(len(neg_index), len(pos_index))

        dice_neg = (p_sum == 0).float()
        dice_pos = 2 * (p * t).sum(-1) / ((p + t).sum(-1))

        dice_neg = dice_neg[neg_index]
        dice_pos = dice_pos[pos_index]
        dice = torch.cat([dice_pos, dice_neg])

        dice_neg = np.nan_to_num(dice_neg.mean().item(), 0)
        dice_pos = np.nan_to_num(dice_pos.mean().item(), 0)
        dice = dice.mean().item()

        num_neg = len(neg_index)
        num_pos = len(pos_index)

    return dice, dice_neg, dice_pos, num_neg, num_pos


def mIoU(results, masks, threshold=0.5, eps=1):
    """
    :param results: b w h
    :param masks: b w h
    :param threshold: 0.5
    :return:
    """
    results = (results > threshold).float()
    b = len(results)
    results = results.view(b, -1).detach().cpu().numpy()
    masks = masks.view(b, -1).detach().cpu().numpy()
    intersection = (masks * results).sum(1)
    union = results.sum(1) + masks.sum(1) - intersection
    union = torch.from_numpy(union)
    intersection = torch.from_numpy(intersection)
    intersection = torch.clamp(intersection, 0, 1e8)
    union = torch.clamp(union, 0, 1e8)
    ious = ((intersection + eps) / (union + eps)).mean()
    return ious


def fbeta_score_threshold_matrix(y_true, y_pred, threshold, beta=2, eps=1e-9):
    beta2 = beta ** 2
    threshold = torch.from_numpy(threshold)
    threshold = threshold.unsqueeze(0).repeat(len(y_pred), 1).float()
    y_pred = (y_pred > threshold).float()
    y_true = y_true.float()

    true_positive = (y_pred * y_true).sum(dim=1)
    precision = true_positive.div(y_pred.sum(dim=1).add(eps))
    recall = true_positive.div(y_true.sum(dim=1).add(eps))

    return torch.mean(
        (precision * recall).div(precision.mul(beta2) + recall + eps).mul(1 + beta2)
    ).item()


def fbeta_score(y_true, y_pred, beta, threshold, eps=1e-9):
    beta2 = beta ** 2

    y_pred = torch.ge(y_pred.float(), threshold).float()
    y_true = y_true.float()

    true_positive = (y_pred * y_true).sum(dim=1)
    precision = true_positive.div(y_pred.sum(dim=1).add(eps))
    recall = true_positive.div(y_true.sum(dim=1).add(eps))

    return torch.mean(
        (precision * recall).div(precision.mul(beta2) + recall + eps).mul(1 + beta2)
    ).item()


def create_class_weight(labels_dict, mu=0.5):
    total = np.sum(np.array(list(labels_dict.values())))
    keys = labels_dict.keys()
    class_weight = dict()
    class_weight_log = dict()
    for key in keys:
        score = total / float(labels_dict[key])
        score_log = math.log(mu * total / float(labels_dict[key]))
        class_weight[key] = round(score, 2) if score > 1.0 else round(1.0, 2)
        class_weight_log[key] = (
            round(score_log, 2) if score_log > 1.0 else round(1.0, 2)
        )

    return class_weight, class_weight_log


def get_weight(targets):
    dict_label = {}
    for Target in targets:
        labels = Target.split(" ")
        labels = [int(l) for l in labels]
        for label in labels:
            if label in dict_label.keys():
                dict_label[label] += 1
            else:
                dict_label[label] = 1
    # n_labels = sum(list(dict_label.values()))
    Target2 = [sorted([int(i) for i in t.split(" ")]) for t in targets]
    class_weight, class_weight_log = create_class_weight(dict_label)
    # prob_dict, prob_dict_bal = cls_wts(dict_label, n_labels)
    weights = []
    for t in Target2:
        w = 0
        for t_ in t:
            w += class_weight_log[t_]
        weights.append(w)
    return weights