Pytorch 实现focal_loss 多类别和二分类示例

时间:2021-09-16 13:48:34

我就废话不多说了,直接上代码吧!

?
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
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
 
 
# 支持多分类和二分类
class FocalLoss(nn.Module):
  """
  This is a implementation of Focal Loss with smooth label cross entropy supported which is proposed in
  'Focal Loss for Dense Object Detection. (https://arxiv.org/abs/1708.02002)'
    Focal_Loss= -1*alpha*(1-pt)^gamma*log(pt)
  :param num_class:
  :param alpha: (tensor) 3D or 4D the scalar factor for this criterion
  :param gamma: (float,double) gamma > 0 reduces the relative loss for well-classified examples (p>0.5) putting more
          focus on hard misclassified example
  :param smooth: (float,double) smooth value when cross entropy
  :param balance_index: (int) balance class index, should be specific when alpha is float
  :param size_average: (bool, optional) By default, the losses are averaged over each loss element in the batch.
  """
 
  def __init__(self, num_class, alpha=None, gamma=2, balance_index=-1, smooth=None, size_average=True):
    super(FocalLoss, self).__init__()
    self.num_class = num_class
    self.alpha = alpha
    self.gamma = gamma
    self.smooth = smooth
    self.size_average = size_average
 
    if self.alpha is None:
      self.alpha = torch.ones(self.num_class, 1)
    elif isinstance(self.alpha, (list, np.ndarray)):
      assert len(self.alpha) == self.num_class
      self.alpha = torch.FloatTensor(alpha).view(self.num_class, 1)
      self.alpha = self.alpha / self.alpha.sum()
    elif isinstance(self.alpha, float):
      alpha = torch.ones(self.num_class, 1)
      alpha = alpha * (1 - self.alpha)
      alpha[balance_index] = self.alpha
      self.alpha = alpha
    else:
      raise TypeError('Not support alpha type')
 
    if self.smooth is not None:
      if self.smooth < 0 or self.smooth > 1.0:
        raise ValueError('smooth value should be in [0,1]')
 
  def forward(self, input, target):
    logit = F.softmax(input, dim=1)
 
    if logit.dim() > 2:
      # N,C,d1,d2 -> N,C,m (m=d1*d2*...)
      logit = logit.view(logit.size(0), logit.size(1), -1)
      logit = logit.permute(0, 2, 1).contiguous()
      logit = logit.view(-1, logit.size(-1))
    target = target.view(-1, 1)
 
    # N = input.size(0)
    # alpha = torch.ones(N, self.num_class)
    # alpha = alpha * (1 - self.alpha)
    # alpha = alpha.scatter_(1, target.long(), self.alpha)
    epsilon = 1e-10
    alpha = self.alpha
    if alpha.device != input.device:
      alpha = alpha.to(input.device)
 
    idx = target.cpu().long()
    one_hot_key = torch.FloatTensor(target.size(0), self.num_class).zero_()
    one_hot_key = one_hot_key.scatter_(1, idx, 1)
    if one_hot_key.device != logit.device:
      one_hot_key = one_hot_key.to(logit.device)
 
    if self.smooth:
      one_hot_key = torch.clamp(
        one_hot_key, self.smooth, 1.0 - self.smooth)
    pt = (one_hot_key * logit).sum(1) + epsilon
    logpt = pt.log()
 
    gamma = self.gamma
 
    alpha = alpha[idx]
    loss = -1 * alpha * torch.pow((1 - pt), gamma) * logpt
 
    if self.size_average:
      loss = loss.mean()
    else:
      loss = loss.sum()
    return loss
 
 
 
class BCEFocalLoss(torch.nn.Module):
  """
  二分类的Focalloss alpha 固定
  """
  def __init__(self, gamma=2, alpha=0.25, reduction='elementwise_mean'):
    super().__init__()
    self.gamma = gamma
    self.alpha = alpha
    self.reduction = reduction
 
  def forward(self, _input, target):
    pt = torch.sigmoid(_input)
    alpha = self.alpha
    loss = - alpha * (1 - pt) ** self.gamma * target * torch.log(pt) - \
        (1 - alpha) * pt ** self.gamma * (1 - target) * torch.log(1 - pt)
    if self.reduction == 'elementwise_mean':
      loss = torch.mean(loss)
    elif self.reduction == 'sum':
      loss = torch.sum(loss)
    return loss

以上这篇Pytorch 实现focal_loss 多类别和二分类示例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/qq_33278884/article/details/91572173