import torchvision
import torchvision.transforms as transforms

train_set = torchvision.datasets.FashionMNIST(root='./data',train=True,download=True,transform=transforms.Compose([transforms.ToTensor()])
)

train_loader = torch.utils.data.DataLoader(train_set,batch_size=1000,shuffle=True
)

import torch
import torchvision.transforms as T
import numpy as np
from torch.utils.data import DataLoader, random_split,Dataset
import PIL.Image as Image
class MiniImageNet(Dataset):def __init__(self,datapath,train,transform=None):self.images = np.load(datapath+'/miniimagenet.npy')self.transform = transformself.trainset = self.images[:50000,:,:]self.testset = self.images[50000:,:,:]self.train = traindef __len__(self):if self.train==True:return self.trainset.shape[0]else:return self.testset.shape[0]def __getitem__(self,index):if self.train==True:image_index = self.trainset[index]else:image_index = self.testset[index]image_index = Image.fromarray(np.uint8(image_index))if self.transform:image_index = self.transform(image_index)return (image_index,0)

def train_transforms(dataset):if dataset == "cifar10":# "When full-network fine-tuning on CIFAR-10 and CIFAR100, we use the augmentation popularized by Wide Residual# Networks: 4 pixels are reflection padded on each side, and# a 32 × 32 crop is randomly sampled from the padded image or its horizontal flip"return T.Compose([T.RandomCrop(32, padding=4, padding_mode="reflect"),T.RandomHorizontalFlip(),T.ToTensor(),])elif dataset == "mnist" or dataset == "fmnist":return T.ToTensor()

def imagenet_dataloader(batch_size):datapath = 'data'train_ds = MiniImageNet(datapath,train=True,transform=train_transforms("cifar10"))valid_ds = MiniImageNet(datapath,train=True,transform=test_transforms("cifar10"))test_ds = MiniImageNet(datapath,train=False,transform=test_transforms("cifar10"))train_size = int(0.9 * len(train_ds))#因为对训练集进行了数据增强，而验证集没有使用，所以要分开写train_ds, _ = random_split(train_ds,[train_size, len(train_ds) - train_size],generator=torch.Generator().manual_seed(0),)_, valid_ds = random_split(valid_ds,[train_size, len(valid_ds) - train_size],generator=torch.Generator().manual_seed(0),)train_dl = DataLoader(train_ds, shuffle=True, batch_size=batch_size, num_workers=8)valid_dl = DataLoader(valid_ds, batch_size=batch_size, num_workers=8)test_dl = DataLoader(test_ds, batch_size=batch_size, num_workers=8)return train_dl, valid_dl, test_dl