Pytorch中transforms.RandomResizedCrop()等图像操作
先看代码：
transforms.Compose([transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
具体是对图像进行各种转换操作，并用函数compose将这些转换操作组合起来；
接下来看实例：
先读取一张图片：
from PIL import Image
img = Image.open("./tulip.jpg")
transforms.RandomResizedCrop(224) 将给定图像随机裁剪为不同的大小和宽高比，然后缩放所裁剪得到的图像为制定的大小；（即先随机采集，然后对裁剪得到的图像缩放为同一大小）
默认scale=(0.08, 1.0)
img = Image.open("./demo.jpg")
print("原图大小：",img.size)
data1 = transforms.RandomResizedCrop(224)(img)
print("随机裁剪后的大小:",data1.size)
data2 = transforms.RandomResizedCrop(224)(img)
data3 = transforms.RandomResizedCrop(224)(img)
plt.subplot(2,2,1),plt.imshow(img),plt.title("原图")
plt.subplot(2,2,2),plt.imshow(data1),plt.title("转换后的图1")
plt.subplot(2,2,3),plt.imshow(data2),plt.title("转换后的图2")
plt.subplot(2,2,4),plt.imshow(data3),plt.title("转换后的图3")
plt.show()
结果为：
原图大小： (500, 721)
随机裁剪后的大小: (224, 224)
该操作的含义在于：即使只是该物体的一部分，我们也认为这是该类物体；
transforms.RandomHorizontalFlip() 以给定的概率随机水平旋转给定的PIL的图像，默认为0.5；
img = Image.open("./demo.jpg")
img1 = transforms.RandomHorizontalFlip()(img)
img2 = transforms.RandomHorizontalFlip()(img)
img3 = transforms.RandomHorizontalFlip()(img)
plt.subplot(2,2,1),plt.imshow(img),plt.title("原图")
plt.subplot(2,2,2), plt.imshow(img1), plt.title("变换后的图1")
plt.subplot(2,2,3), plt.imshow(img2), plt.title("变换后的图2")
plt.subplot(2,2,4), plt.imshow(img3), plt.title("变换后的图3")
plt.show()
transforms.ToTensor() 将给定图像转为Tensor
img = Image.open("./demo.jpg")
img = transforms.ToTensor()(img)
print(img)
输出为：
tensor([[[0.4549, 0.4549, 0.4471, ..., 0.5216, 0.5294, 0.5294],
[0.4510, 0.4510, 0.4431, ..., 0.5216, 0.5255, 0.5255],
[0.4471, 0.4431, 0.4392, ..., 0.5176, 0.5255, 0.5216],
...,
[0.5529, 0.5333, 0.5059, ..., 0.7922, 0.7922, 0.7922],
[0.5647, 0.5451, 0.5176, ..., 0.7922, 0.7922, 0.7922],
[0.5882, 0.5725, 0.5451, ..., 0.7843, 0.7843, 0.7843]],
[[0.4980, 0.4980, 0.4902, ..., 0.5059, 0.5137, 0.5137],
[0.4941, 0.4941, 0.4863, ..., 0.5059, 0.5098, 0.5098],
[0.4902, 0.4863, 0.4824, ..., 0.5020, 0.5098, 0.5059],
...,
[0.5059, 0.4863, 0.4588, ..., 0.7373, 0.7373, 0.7373],
[0.5176, 0.4980, 0.4706, ..., 0.7373, 0.7373, 0.7373],
[0.5412, 0.5255, 0.4980, ..., 0.7451, 0.7451, 0.7451]],
[[0.5137, 0.5137, 0.5059, ..., 0.5020, 0.5098, 0.5098],
[0.5098, 0.5098, 0.5020, ..., 0.5020, 0.5059, 0.5059],
[0.5059, 0.5020, 0.4980, ..., 0.4980, 0.5059, 0.5020],
...,
[0.4431, 0.4235, 0.3961, ..., 0.7373, 0.7373, 0.7373],
[0.4549, 0.4353, 0.4078, ..., 0.7373, 0.7373, 0.7373],
[0.4941, 0.4784, 0.4510, ..., 0.7490, 0.7490, 0.7490]]])
transforms.Normalize(） 归一化处理
img = Image.open("./demo.jpg")
img = transforms.ToTensor()(img)
img = transforms.Normalize(mean=[0.5,0.5,0.5], std=[0.5,0.5,0.5])(img)
print(img)
输出：
tensor([[[-0.0902, -0.0902, -0.1059, ..., 0.0431, 0.0588, 0.0588],
[-0.0980, -0.0980, -0.1137, ..., 0.0431, 0.0510, 0.0510],
[-0.1059, -0.1137, -0.1216, ..., 0.0353, 0.0510, 0.0431],
...,
[ 0.1059, 0.0667, 0.0118, ..., 0.5843, 0.5843, 0.5843],
[ 0.1294, 0.0902, 0.0353, ..., 0.5843, 0.5843, 0.5843],
[ 0.1765, 0.1451, 0.0902, ..., 0.5686, 0.5686, 0.5686]],
[[-0.0039, -0.0039, -0.0196, ..., 0.0118, 0.0275, 0.0275],
[-0.0118, -0.0118, -0.0275, ..., 0.0118, 0.0196, 0.0196],
[-0.0196, -0.0275, -0.0353, ..., 0.0039, 0.0196, 0.0118],
...,
[ 0.0118, -0.0275, -0.0824, ..., 0.4745, 0.4745, 0.4745],
[ 0.0353, -0.0039, -0.0588, ..., 0.4745, 0.4745, 0.4745],
[ 0.0824, 0.0510, -0.0039, ..., 0.4902, 0.4902, 0.4902]],
[[ 0.0275, 0.0275, 0.0118, ..., 0.0039, 0.0196, 0.0196],
[ 0.0196, 0.0196, 0.0039, ..., 0.0039, 0.0118, 0.0118],
[ 0.0118, 0.0039, -0.0039, ..., -0.0039, 0.0118, 0.0039],
...,
[-0.1137, -0.1529, -0.2078, ..., 0.4745, 0.4745, 0.4745],
[-0.0902, -0.1294, -0.1843, ..., 0.4745, 0.4745, 0.4745],
[-0.0118, -0.0431, -0.0980, ..., 0.4980, 0.4980, 0.4980]]])