from torchvision import transforms
unloader = transforms.ToPILImage()
image = original_tensor.cpu().clone() ?# clone the tensor
image = image.squeeze(0) ?# remove the fake batch dimension
image = unloader(image)
image.save('example.jpg')

def tensor2im(input_image, imtype=np.uint8):
    """"
    Parameters:
        input_image (tensor) --  輸入的tensor，維度為CHW，注意這里沒有batch size的維度
        imtype (type)        --  轉(zhuǎn)換后的numpy的數(shù)據(jù)類型
    """
    mean = [0.485, 0.456, 0.406] # dataLoader中設(shè)置的mean參數(shù)，需要從dataloader中拷貝過來
    std = [0.229, 0.224, 0.225]  # dataLoader中設(shè)置的std參數(shù)，需要從dataloader中拷貝過來
    if not isinstance(input_image, np.ndarray):
        if isinstance(input_image, torch.Tensor): # 如果傳入的圖片類型為torch.Tensor，則讀取其數(shù)據(jù)進(jìn)行下面的處理
            image_tensor = input_image.data
        else:
            return input_image
        image_numpy = image_tensor.cpu().float().numpy()  # convert it into a numpy array
        if image_numpy.shape[0] == 1:  # grayscale to RGB
            image_numpy = np.tile(image_numpy, (3, 1, 1))
        for i in range(len(mean)): # 反標(biāo)準(zhǔn)化，乘以方差，加上均值
            image_numpy[i] = image_numpy[i] * std[i] + mean[i]
        image_numpy = image_numpy * 255 #反ToTensor(),從[0,1]轉(zhuǎn)為[0,255]
        image_numpy = np.transpose(image_numpy, (1, 2, 0))  # 從(channels, height, width)變?yōu)?height, width, channels)
    else:  # 如果傳入的是numpy數(shù)組,則不做處理
        image_numpy = input_image
    return image_numpy.astype(imtype)