(predicted == labels).sum().item()

import torch
import numpy as np 

data1 = np.array([
    [1,2,3],
    [2,3,4]
])
data1_torch = torch.from_numpy(data1)

data2 = np.array([
    [1,2,3],
    [2,3,4]
])
data2_torch = torch.from_numpy(data2)

p = (data1_torch == data2_torch) #对比后相同的值会为1，不同则会为0
print p
print type(p)

d1 = p.sum() #将所有的值相加，得到的仍是tensor类别的int值
print d1 
print type(d1)

d2 = d1.item() #转成python数字
print d2 
print type(d2)

输出：

tensor([[1, 1, 1],
        [1, 1, 1]], dtype=torch.uint8)
<class 'torch.Tensor'>
tensor(6)
<class 'torch.Tensor'>
6
<type 'int'>

从而理解： 

 correct += (predicted == labels).sum().item()  # 如果预测结果和真实值相等则计数 +1