BCELoss就时predict对应位置点求熵,相加,再求平均
证明:
#%%
import torch
import torch.nn as nn
#%% md
BCELoss损失函数
#%% 产生predict,target
N = 2
C = 1
H = 2
W = 2
predict = torch.arange(N*C*H*W,dtype=torch.float32).view([N,C,W,H])
predict = torch.sigmoid(predict) #sigmoid!!!
target = torch.arange(N*C*H*W,dtype=torch.float32).view([N,C,W,H]) #sigmoid!!!
target = torch.sigmoid(target)
print('predict:',predict)
print('target:',target)
#%% 利用公式计算BCELoss
sumAll = 0
for n in range(N):
for c in range(C):
for h in range(H):
for w in range(W):
sumAll += target[n,c,h,w]*torch.log(predict[n,c,h,w])+(1-target[n,c,h,w])*torch.log(1-predict[n,c,h,w])
print('公式计算结果:BCELoss:',-sumAll/(N*C*H*W))
#%% 利用pytorch计算BCELoss
loss = torch.nn.BCELoss()
print('Pytorch计算结果:BCELoss:',loss(predict,target))
结果:
公式计算结果:BCELoss: tensor(0.2483)
Pytorch计算结果:BCELoss: tensor(0.2483)