pytorch加载自己的数据，自定义方式

import torch.nn as nn
import torch
import torch.nn.functional as F
from torch.autograd import Variable
import numpy as np
from numpy import *
import random
import matplotlib.pyplot as plt
from torch.utils.data import DataLoader,Dataset
from torchvision import datasets, transforms
import os
from PIL import Image
from torchvision.transforms import Resize

    """默认图像数据目录结构
    root
    .
    ├──dog
    |   ├──001.png
    |   ├──002.png
    |   └──...
    └──cat  
    |   ├──001.png
    |   ├──002.png
    |   └──...
    └──...
    """
def find_class(path):
	classes = os.listdir(path)
	class_to_idx = {
    
    }
	for i ,clss in enumerate(classes):
		class_to_idx[clss] = i
	return classes,class_to_idx

def make_data(dir_path,class_to_idx):
	img = []
	for i in class_to_idx:
		im = dir_path + '/' + i
		for j in os.listdir(im):
			img.append([im + '/' +j,class_to_idx[i]])
	return img

class Data_jiazai(Dataset):
	"""docstring for Data_jiazai"""
	def __init__(self, dir_path,transform=None):
		super(Data_jiazai, self).__init__()
		self.dir_path = dir_path
		self.transform = transform
		classes,class_to_idx = find_class(dir_path)
		self.img = make_data(dir_path,class_to_idx)
	def __getitem__(self,index):
		# for names in (classes):
		image_path,label =  self.img[index][0],self.img[index][1]
		image = Image.open(image_path)

		image = self.transform(image)
		return image,label

	def __len__(self):
		return len(self.img)

transform2 = transforms.Compose([transforms.CenterCrop(size=28),transforms.Resize((200, 200)),transforms.ToTensor(), transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])])
data = Data_jiazai('./catdog/train/',transform=transform2)
train_data = DataLoader(data,batch_size=batch_size,shuffle=True)#使用DataLoader加载数据
for i,(j,k) in enumerate(dataloader):
 	print(i,j,k)