转自https://blog.csdn.net/tsq292978891/article/details/79403617

备份自用，不喜勿喷

torchvision在pypi上的文档介绍 PyTorch 0.3.0 中文文档

简介: torchvision包是服务于pytorch深度学习框架的,用来生成图片,视频数据集,和一些流行的模型类和预训练模型.  torchvision由以下四个部分组成:  1. torchvision.datasets : Data loaders for popular vision datasets  2. torchvision.models : Definitions for popular model architectures, such as AlexNet, VGG, and ResNet and pre-trained models.  3. torchvision.transforms : Common image transformations such as random crop, rotations etc.  4. torchvision.utils : Useful stuff such as saving tensor (3 x H x W) as image to disk, given a mini-batch creating a grid of images, etc.

下面分别介绍

第一部分: torchvision.datasets torchvision.datasets是继承torch.utils.data.Dataset的子类. 因此,可以使用torch.utils.data.DataLoader对它们进行多线程处理(python multiprocessing)  比如:  

torchvision.datasets可能需要transform和target_transform参数,关于二者的解释如下:

transform - a function that takes in an image and returns a transformed version common stuff like ToTensor, RandomCrop, etc. These can be composed together with transforms.Compose (see transforms section below) 输入原始图片，返回转换后的图片 target_transform - a function that takes in the target and transforms it. For example, take in the caption string and return a tensor of word indices. 输入为 target, 返回转换后的 target torchvision.datasets包括以下内容:  MNIST  COCO (Captioning and Detection)  LSUN Classification  ImageFolder  Imagenet-12  CIFAR10 and CIFAR100  STL10  SVHN  PhotoTour  其中，ImageFolder是一种data loader．图片以下面的方式存放：  root/dog/xxx.png  root/dog/xxy.png  root/dog/xxz.png

root/cat/123.png  root/cat/nsdf3.png  root/cat/asd932_.png # 不同类别的图片放在各自的文件夹下

dset.ImageFolder(root=”root folder path”, [transform, target_transform])  然后，ImageFolder类有下面三个成员属性：  (1) self.classes - The class names as a list (类别名字列表)  (2) self.class_to_idx - Corresponding class indices　(类别对应的序号)  (3) self.imgs - The list of (image path, class-index) tuples　(图片路径＋类别序号组成的元组)

第二部分：torchvision.models torchvision.models包含下列模型的定义：

AlexNet: AlexNet variant from the “One weird trick” paper. VGG: VGG-11, VGG-13, VGG-16, VGG-19 (with and without batch normalization) ResNet: ResNet-18, ResNet-34, ResNet-50, ResNet-101, ResNet-152 SqueezeNet: SqueezeNet 1.0, and SqueezeNet 1.1 使用方式１：构建一个模型，随机初始化参数

import torchvision.models as models resnet18 = models.resnet18() alexnet = models.alexnet() vgg16 = models.vgg16() squeezenet = models.squeezenet1_0() 1 2 3 4 5 使用方式２：构建一个模型，使用预训练的模型进行参数初始化．  We provide pre-trained models for the ResNet variants, SqueezeNet 1.0 and 1.1, and AlexNet, using the PyTorch model zoo. These can be constructed by passing pretrained=True．  有预训练模型的网络有：ResNet variants, SqueezeNet 1.0 and 1.1, and AlexNet．构建预训练模型使用了torch.utils.model_zoo．设置pretrained=True

import torchvision.models as models resnet18 = models.resnet18(pretrained=True) alexnet = models.alexnet(pretrained=True) squeezenet = models.squeezenet1_0(pretrained=True) 1 2 3 4 注：这些pre-trained models要求输入图片格式如下：  1. 像素值范围[0, 1]，normalized，mean=[0.485, 0.456, 0.406]，std=[0.229, 0.224, 0.225]  2. mini-batches　RGB images，shape (3 x H x W)，H和W至少224. (输入图片: NCHW)

imagenet推荐的normalization例子:

Data loading code traindir = os.path.join(args.data, 'train') valdir = os.path.join(args.data, 'val') normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],                                  std=[0.229, 0.224, 0.225])

train_loader = torch.utils.data.DataLoader(     datasets.ImageFolder(traindir, transforms.Compose([         transforms.RandomSizedCrop(224),         transforms.RandomHorizontalFlip(),         transforms.ToTensor(),         normalize,     ])),     batch_size=args.batch_size, shuffle=True,     num_workers=args.workers, pin_memory=True)

val_loader = torch.utils.data.DataLoader(     datasets.ImageFolder(valdir, transforms.Compose([         transforms.Scale(256),         transforms.CenterCrop(224),         transforms.ToTensor(),         normalize,     ])),     batch_size=args.batch_size, shuffle=False,     num_workers=args.workers, pin_memory=True) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 第三部分: torchvision.transforms torchvision.transforms包含了常见的图像变化(预处理)操作.这些变化可以用torchvision.transforms.Compose链接在一起.  torchvision.transforms中的变化, 可以分为以下几类:  一: Transforms on PIL.Image  1. Scale(size, interpolation=Image.BILINEAR)  2. CenterCrop(size) - center-crops the image to the given size  3. RandomCrop(size, padding=0)  4. RandomHorizontalFlip()  5. RandomSizedCrop(size, interpolation=Image.BILINEAR)  6. Pad(padding, fill=0)

二: Transforms on torch.*Tensor  1. Normalize(mean, std)  作用: Given mean: (R, G, B) and std: (R, G, B), will normalize each channel of the torch.*Tensor, i.e. channel = (channel - mean) / std

三: Conversion Transforms 数据格式转换操作  1. ToTensor()  作用: Converts a PIL.Image (RGB) or numpy.ndarray (H x W x C) in the range [0, 255] to a torch.FloatTensor of shape (C x H x W) in the range [0.0, 1.0]

四: Generic Transforms 一般的变化操作  1. Lambda(lambda) # 自己定义一个python lambda表达式, applies it to the input img and returns it.  举例: transforms.Lambda(lambda x: x.add(10))　 # 将每个像素值加10

第四部分: torchvision.utils utils嘛, 就是一些工具. 好像目前只有两个.  1. torchvision.utils.make_grid(tensor, nrow=8, padding=2, normalize=False, range=None, scale_each=False)  作用: 输入4D mini-batch Tensor of shape (B x C x H x W)或者a list of images all of the same size, 然后用这些图片生成一个大的图片.(图片中每格子为单张图片)  normalize=True will shift the image to the range (0, 1), by subtracting the minimum and dividing by the maximum pixel value.  if range=(min, max) where min and max are numbers, then these numbers are used to normalize the image.  scale_each=True will scale each image in the batch of images separately rather than computing the (min, max) over all images.

一个例子:

torchvision.utils.save_image(tensor, filename, nrow=8, padding=2, normalize=False, range=None, scale_each=False)  作用: 将输入的Tensor保存为image file. 如果输入的是mini-batch tensor, 则会保存a grid of images. All options after filename are passed through to make_grid.