dataset.ImageFolder(root="datapath", transfroms.ToTensor())

#读取图片　这里是灰度图　
 for item in all_path:
  img = cv2.imread(item[1],0)
  img = cv2.resize(img,(28,28))
  arr = np.asarray(img,dtype="float32")
  data_x[i ,:,:,:] = arr
  i+=1
  data_y.append(int(item[0]))
  
 data_x = data_x / 255
 data_y = np.asarray(data_y)

 data_x = torch.from_numpy(data_x)
 data_y = torch.from_numpy(data_y)

 dataset = dataf.TensorDataset(data_x,data_y)
 loader = dataf.DataLoader(dataset, batch_size=batchsize, shuffle=True)

# coding=utf-8
import os
import cv2
import numpy as np
import random
 
import torch
import torch.nn as nn
import torch.utils.data as dataf
from torch.autograd import Variable
import torch.nn.functional as F
import torch.optim as optim
 
#训练参数
cuda = False
train_epoch = 20
train_lr = 0.01
train_momentum = 0.5
batchsize = 5
 
 
#测试训练集路径
test_path = "/home/test/"
train_path = "/home/train/"
 
#路径数据
all_path =[]
 
def load_data(data_path):
 signal = os.listdir(data_path)
 for fsingal in signal: 
  filepath = data_path+fsingal
  filename = os.listdir(filepath)
  for fname in filename:
   ffpath = filepath+"/"+fname
   path = [fsingal,ffpath]
   all_path.append(path)
   
#设立数据集多大
 count = len(all_path)
 data_x = np.empty((count,1,28,28),dtype="float32")
 data_y = []
#打乱顺序
 random.shuffle(all_path)
 i=0;
 
#读取图片　这里是灰度图　最后结果是i*i*i*i
#分别表示：batch大小　，　通道数，　像素矩阵
 for item in all_path:
  img = cv2.imread(item[1],0)
  img = cv2.resize(img,(28,28))
  arr = np.asarray(img,dtype="float32")
  data_x[i ,:,:,:] = arr
  i+=1
  data_y.append(int(item[0]))
  
 data_x = data_x / 255
 data_y = np.asarray(data_y)
#  lener = len(all_path)
 data_x = torch.from_numpy(data_x)
 data_y = torch.from_numpy(data_y)
 dataset = dataf.TensorDataset(data_x,data_y)
 
 loader = dataf.DataLoader(dataset, batch_size=batchsize, shuffle=True)
  
 return loader
#  print data_y
 
 
 
train_load = load_data(train_path)
test_load = load_data(test_path)
 
class L5_NET(nn.Module):
 def __init__(self):
  super(L5_NET ,self).__init__();
  #第一层输入1，20个卷积核　每个5*5
  self.conv1 = nn.Conv2d(1 , 20 , kernel_size=5)
  #第二层输入20，30个卷积核　每个5*5
  self.conv2 = nn.Conv2d(20 , 30 , kernel_size=5)
  #drop函数
  self.conv2_drop = nn.Dropout2d()
  #全链接层１，展开30*4*4，连接层50个神经元
  self.fc1 = nn.Linear(30*4*4,50)
  #全链接层１，50-4 ,4为最后的输出分类
  self.fc2 = nn.Linear(50,4)
 
 #前向传播
 def forward(self,x):
  #池化层１ 对于第一层卷积池化，池化核2*2
  x = F.relu(F.max_pool2d( self.conv1(x)  ,2 ) )
  #池化层2 对于第二层卷积池化，池化核2*2
  x = F.relu(F.max_pool2d( self.conv2_drop( self.conv2(x) ) , 2 ) )
  #平铺轴30*4*4个神经元
  x = x.view(-1 , 30*4*4)
  #全链接１
  x = F.relu( self.fc1(x) )
  #dropout链接
  x = F.dropout(x , training= self.training)
  #全链接w
  x = self.fc2(x)
  #softmax链接返回结果
  return F.log_softmax(x)
 
model = L5_NET()
if cuda :
 model.cuda()
  
 
optimizer = optim.SGD(model.parameters()  , lr =train_lr , momentum = train_momentum )
 
#预测函数
def train(epoch):
 model.train()
 for batch_idx, (data, target) in enumerate(train_load):
  if cuda:
   data, target = data.cuda(), target.cuda()
  data, target = Variable(data), Variable(target)
  #求导
  optimizer.zero_grad()
  #训练模型，输出结果
  output = model(data)
  #在数据集上预测loss
  loss = F.nll_loss(output, target)
  #反向传播调整参数pytorch直接可以用loss
  loss.backward()
  #SGD刷新进步
  optimizer.step()
  #实时输出
  if batch_idx % 10 == 0:
   print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
    epoch, batch_idx * len(data), len(train_load.dataset),
    100. * batch_idx / len(train_load), loss.data[0]))
#    
   
#测试函数
def test(epoch):
 model.eval()
 test_loss = 0
 correct = 0
 for data, target in test_load:
  
  if cuda:
   data, target = data.cuda(), target.cuda()
   
  data, target = Variable(data, volatile=True), Variable(target)
  #在测试集上预测
  output = model(data)
  #计算在测试集上的loss
  test_loss += F.nll_loss(output, target).data[0]
  #获得预测的结果
  pred = output.data.max(1)[1] # get the index of the max log-probability
  #如果正确，correct+1
  correct += pred.eq(target.data).cpu().sum()
 
 #loss计算
 test_loss = test_loss
 test_loss /= len(test_load)
 #输出结果
 print('\nThe {} epoch result : Average loss: {:.6f}, Accuracy: {}/{} ({:.2f}%)\n'.format(
  epoch,test_loss, correct, len(test_load.dataset),
  100. * correct / len(test_load.dataset)))
 
for epoch in range(1, train_epoch+ 1):
 train(epoch)
 test(epoch)