from torchvision import transforms, datasets as dsfrom torch.utils.data import DataLoaderimport matplotlib.pyplot as pltimport numpy as np#transform = transforms.Compose是把一系列圖片操作組合起來，比如減去像素均值等。#DataLoader讀入的數(shù)據(jù)類型是PIL.Image#這里對(duì)圖片不做任何處理，僅僅是把PIL.Image轉(zhuǎn)換為torch.FloatTensor，從而可以被pytorch計(jì)算transform = transforms.Compose( [ transforms.ToTensor() ])

train_set = ds.CIFAR100(root='.', train=True, transform=transform, target_transform=None, download=True)

data_loader = DataLoader(dataset=train_set,  batch_size=1,  shuffle=False,  num_workers=2)# # 生成torch.utils.data.DataLoaderIter# # 不過DataLoaderIter它會(huì)被DataLoader自動(dòng)創(chuàng)建并且調(diào)用，我們用不到# data_iter = iter(data_loader)# images, labels = next(data_iter)

to_pil_image = transforms.ToPILImage()cnt = 0for image,label in data_loader: if cnt>=3: # 只顯示3張圖片 break print(label) # 顯示label # 方法1：Image.show() # transforms.ToPILImage()中有一句 # npimg = np.transpose(pic.numpy(), (1, 2, 0)) # 因此pic只能是3-D Tensor，所以要用image[0]消去batch那一維 img = to_pil_image(image[0]) img.show() # 方法2：plt.imshow(ndarray) img = image[0] # plt.imshow()只能接受3-D Tensor，所以也要用image[0]消去batch那一維 img = img.numpy() # FloatTensor轉(zhuǎn)為ndarray img = np.transpose(img, (1,2,0)) # 把channel那一維放到最后 # 顯示圖片 plt.imshow(img) plt.show() cnt += 1