Pytorch深度学习------Torchvision中Transforms的使用（ToTensor，Normalize，Resize ，Compose，RandomCrop）

一、transforms的作用

transforms主要用于对图像进行变换，是常见的图像预处理方法。

二、transforms结构

通过pycharm中在该transforms.py下的structure进行查看

三、transforms的使用

首先引入transforms所在的模块

from torchvision import transforms

其次由上面的截图可以知道transforms就是一个.py文件，里面有很多的类以及方法,所以要先使用就必须通过.的形式去调用，如果要调用类，那就需要实例化，实例化后就是一个对象，对象就可以使用该类下的方法。

1、ToTensor类

class ToTensor:
    """Convert a PIL Image or ndarray to tensor and scale the values accordingly.

将PIL Image or ndarray类型转化为tensor 类型

from PIL import Image
from torchvision import transforms

img_path = "hymenoptera_data/train/ants_image/20935278_9190345f6b.jpg"
# 打开一张图片
img = Image.open(img_path)
# 使用transforms对图像进行变换
# 实例化totensor对象
to_tens = transforms.ToTensor()
# 将pil变成Tensor类型的图片
tens_img = to_tens(img) # 自动调用call函数
print(tens_img)

打印结果：

tensor([[[0.6784, 0.6863, 0.6863,  ..., 0.2431, 0.2471, 0.2549],
         [0.6824, 0.6824, 0.6824,  ..., 0.2078, 0.2118, 0.2157],
         [0.6745, 0.6745, 0.6784,  ..., 0.1882, 0.1882, 0.1843],
         ...,
         [0.5451, 0.5373, 0.5294,  ..., 0.1216, 0.1216, 0.1216],
         [0.5412, 0.5333, 0.5294,  ..., 0.1294, 0.1294, 0.1333],
         [0.5333, 0.5294, 0.5294,  ..., 0.1137, 0.1216, 0.1255]],

        [[0.0588, 0.0588, 0.0588,  ..., 0.5176, 0.5216, 0.5294],
         [0.0549, 0.0549, 0.0588,  ..., 0.4863, 0.4902, 0.4941],
         [0.0510, 0.0510, 0.0549,  ..., 0.4588, 0.4588, 0.4549],
         ...,
         [0.0353, 0.0353, 0.0314,  ..., 0.2902, 0.2941, 0.3059],
         [0.0314, 0.0314, 0.0314,  ..., 0.2902, 0.3020, 0.3098],
         [0.0196, 0.0235, 0.0314,  ..., 0.3059, 0.3137, 0.3176]],

        [[0.5961, 0.5922, 0.5765,  ..., 0.3216, 0.3255, 0.3333],
         [0.5804, 0.5725, 0.5647,  ..., 0.2275, 0.2314, 0.2353],
         [0.5569, 0.5529, 0.5490,  ..., 0.1333, 0.1333, 0.1294],
         ...,
         [0.3725, 0.3373, 0.3020,  ..., 0.0824, 0.0863, 0.0941],
         [0.3765, 0.3333, 0.3020,  ..., 0.0627, 0.0706, 0.0784],
         [0.3843, 0.3451, 0.3098,  ..., 0.0863, 0.0941, 0.0980]]])

Process finished with exit code 0

tensor类型的数据包含哪些属性？

读取PIL Image类型用Image类，读取ndarray 类型用opencv，下面使用opencv。

import cv2
from torchvision import transforms

img_path = "hymenoptera_data/train/ants_image/20935278_9190345f6b.jpg"
# 打开一张图片,并使其类型为ndarray
img = cv2.imread(img_path)
print(type(img)) # 
# 使用transforms对图像进行变换
# 实例化totensor对象
to_tens = transforms.ToTensor()
# 将numpy下的ndarray变成Tensor类型的图片
tens_img = to_tens(img) # 自动调用call函数
print(tens_img)

2、使用上一篇文章中的tensorboard进行查看

打开tensorboad的方法，进入终端tensorboard --logdir=log文件夹名称

from PIL import Image
from torchvision import transforms
from torch.utils.tensorboard import SummaryWriter

img_path = "hymenoptera_data/train/ants_image/20935278_9190345f6b.jpg"
# 打开一张图片
img = Image.open(img_path)
# 使用transforms对图像进行变换
# 实例化totensor对象
to_tens = transforms.ToTensor()
# 将pil变成Tensor类型的图片
tens_img = to_tens(img) # 自动调用call函数
print(tens_img)

# 使用上一篇文章中tensorboard进行查看
writer = SummaryWriter("transforms_logs")
writer.add_image("test_transforms",tens_img) # 标题，图像类型
writer.close()

四、其它常见的transforms

1、Normalize

class Normalize(torch.nn.Module):
    """Normalize a tensor image with mean and standard deviation.
    This transform does not support PIL Image.
    Given mean: ``(mean[1],...,mean[n])`` and std: ``(std[1],..,std[n])`` for ``n``
    channels, this transform will normalize each channel of the input
    ``torch.*Tensor`` i.e.,
    ``output[channel] = (input[channel] - mean[channel]) / std[channel]``

    .. note::
        This transform acts out of place, i.e., it does not mutate the input tensor.

    Args:
        mean (sequence): Sequence of means for each channel.
        std (sequence): Sequence of standard deviations for each channel.
        inplace(bool,optional): Bool to make this operation in-place.

    """

由此可见，该类实例化时需要两个参数，第一个参数mean 代表每个通道上所有数值的平均值，第二个参数std 代表每个通道上所有数值的标准差。而图片的channels查看，可以通过print(img)进行查看mode是几个通道，下面demo的mode是rgba四个通道，故此时的均值和标准差应对应是四列，即会利用这两个参数分别将每层标准化（使数据均值为0，方差为1）后输出。
归一化计算公式：output[channel] = (input[channel] - mean[channel]) / std[channel]
即如果mean和std均为0.5，则代入公式中可以得到：
output[channel] = (input[channel] - 0.5) / 0.5
即：output[channel] = 2*input[channel]-1。

import cv2
from torchvision import transforms
from torch.utils.tensorboard import SummaryWriter
from PIL import Image

writer = SummaryWriter("logs_2")  # 定义存放log文件的文件夹
img = Image.open("image/python_img.png")  # 打开图片
print(img)  # 
# 1. ToTensor
trans_tens = transforms.ToTensor()
img_tens = trans_tens(img) # 将img图片转化tensor类型
writer.add_image("ToTensor",img_tens)
# 2. Normalize 归一化
# 实例化Normalize对象
trans_norm = transforms.Normalize([0.5,0.5,0.5,0.5],[0.5,0.5,0.5,0.5]) # 图片的mode为四个通道RGBA，所以必须四列
# 将tensor类型的图片进行归一化
img_norm = trans_norm(img_tens) # 将Tensor类型图片进行归一化
# 使用tensorboard
writer.add_image("norm",img_norm)  # 进行图片可视化，参数：标题，图像类型，步骤
writer.close()

2、Resize

把输入的图片按照输入的参数值(h,w)重新设定大小。

class Resize(torch.nn.Module):
    """Resize the input image to the given size.
    If the image is torch Tensor, it is expected
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions

    .. warning::
        The output image might be different depending on its type: when downsampling, the interpolation of PIL images
        and tensors is slightly different, because PIL applies antialiasing. This may lead to significant differences
        in the performance of a network. Therefore, it is preferable to train and serve a model with the same input
        types. See also below the ``antialias`` parameter, which can help making the output of PIL images and tensors
        closer.

由

 Args:
        size (sequence or int): Desired output size. If size is a sequence like
            (h, w), output size will be matched to this. If size is an int,
            smaller edge of the image will be matched to this number.
            i.e, if height > width, then image will be rescaled to
            (size * height / width, size).

可知，当如果输入的参数是一个序列，即长和宽两个整数（h,w），则图像会按该长和宽进行resize。
如果输入的参数是一个整数x，将图片短边缩放至x，长宽比保持不变。

import cv2
from torchvision import transforms
from torch.utils.tensorboard import SummaryWriter
from PIL import Image

writer = SummaryWriter("logs_2")  # 定义存放log文件的文件夹
img = Image.open("image/python_img.png")  # 打开图片
print(img)  # 
# 1. ToTensor
trans_tens = transforms.ToTensor()
img_tens = trans_tens(img) # 将img图片转化tensor类型
writer.add_image("ToTensor",img_tens)
# 2. Normalize 归一化
# 实例化Normalize对象
trans_norm = transforms.Normalize([0.5,0.5,0.5,0.5],[0.5,0.5,0.5,0.5]) # 图片的mode为四个通道RGBA，所以必须四列
# 将tensor类型的图片进行归一化
img_norm = trans_norm(img_tens) # 将Tensor类型图片进行归一化
# 使用tensorboard
writer.add_image("norm",img_norm)  # 进行图片可视化，参数：标题，图像类型，步骤
# 3. Resize 把输入的图片按照输入的参数值(h,w)重新设定大小。
trans_resize = transforms.Resize((512,512))
# 将 img大小进行重新设定，并返回pIL类型
img_resize = trans_resize(img)
print(img_resize)  # 
# 将PIL类型转化为Tensor类型
img_resize_tens = trans_tens(img_resize)
# 写入tensorboard中
writer.add_image("resize",img_resize_tens,0)
writer.close()

3、compose

可以理解为多个Transforms操作的组合。即可以一次对输入的图片进行多个Transforms的操作
需要注意：compose中参数是前面的输出作为后面的输入，比如compose中第一个参数的输出是PIL类型，后面的输入也是PIL类型，所以可以直接使用compose，但是如果现在第一个的输出是tensor类型，但是第二个要求的输入是PIL,则会是类型不匹配，所以会报错。

from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms
from PIL import Image

writer = SummaryWriter("logs_2")  # 定义存放log文件的文件夹
img = Image.open("image/python_img.png")  # 打开图片
# 将img图片转化tensor类型,并返回tensor类型
trans_tens = transforms.ToTensor()
print(trans_tens)
# 将 img大小进行重新设定，并返回pIL类型
trans_resize = transforms.Resize(300)
print(trans_resize)
# compose 将ToTensor和Resize进行组合
trans_compose = transforms.Compose([trans_tens,trans_resize])
# 在tensorboard中可视化显示出来
compose_resize = trans_compose(img)
writer.add_image("compose",compose_resize,1)
writer.close()

4、RandomCrop

随机裁剪：对图像进行随机大小裁剪，如果指定数字后就是按照数字大小进行裁剪。

class RandomCrop(torch.nn.Module):
    """Crop the given image at a random location.
    If the image is torch Tensor, it is expected
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions,
    but if non-constant padding is used, the input is expected to have at most 2 leading dimensions

    Args:
        size (sequence or int): Desired output size of the crop. If size is an
            int instead of sequence like (h, w), a square crop (size, size) is
            made. If provided a sequence of length 1, it will be interpreted as (size[0], size[0]).
        padding (int or sequence, optional): Optional padding on each border
            of the image. Default is None. If a single int is provided this
            is used to pad all borders. If sequence of length 2 is provided this is the padding
            on left/right and top/bottom respectively. If a sequence of length 4 is provided
            this is the padding for the left, top, right and bottom borders respectively.

from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms
from PIL import Image

writer = SummaryWriter("logs_2")  # 定义存放log文件的文件夹
img = Image.open("image/python_img.png")  # 打开图片
# 将img图片转化tensor类型,并返回tensor类型
trans_tens = transforms.ToTensor()
print(trans_tens)

# RandomCrop随机裁剪
trans_rand = transforms.RandomCrop(512)  # 裁剪为512*512
# compose 将ToTensor和Resize进行组合
trans_compose = transforms.Compose([trans_rand,trans_tens])
# 随机裁剪10张并在tensorboard中可视化显示出来
for i in range(10):
    img_trans_rand = trans_compose(img)
    print(img_trans_rand)
    writer.add_image("rands",img_trans_rand,i)
writer.close()