Pytorch框架学习记录3——Transform的使用

1. 导入

from torchvision import transforms
1

2.ToTensor的用法

作用：将PIL图像或者numpy转换成tensor格式

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

img_path = "C:\\Users\\hp\\PycharmProjects\pythonProject\\Pytorch_Learning\\flower_data\\train\daisy\\5547758_eea9edfd54_n.jpg"
img = Image.open(img_path)
tensor = transforms.ToTensor()
tensor_img = tensor(img)

writer = SummaryWriter("logs")

writer.add_image("tensor_img", tensor_img, 1)

writer.close()
1
2
3
4
5
6
7
8
9
10
11
12
13
14

3. Normalize的用法

用均值和标准差归一化张量图像，参数：mean, std；输入图像类型为tensor，输出类型为tensor
$$output[channel]=(input[channel]-mean[channel])/std[channel]$$

# Normalize
transform_normal = transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
img_normal = transform_normal(img_tensor)
print(type(img_normal))
writer.add_image("Normalize", img_normal, 1)
1
2
3
4
5

4. RandomCrop 的用法

随机裁剪，参数size表示随机裁剪图片的大小，输入的图像类型为PIL类型，输出的类型也为PIL类型。

下面的这个例子是随机对图像剪裁10次，每次剪裁后的图像大小为32*32

# RandomCrop
transform_randomCrop = transforms.RandomCrop((32, 32))
for i in range(0, 10):
    img_randomCrop = transform_randomCrop(img_PIL)
    # print(type(img_randomCrop))
    img_randomCrop = transform_tensor(img_randomCrop)
    writer.add_image("RandomCrop", img_randomCrop, i)
1
2
3
4
5
6
7

5. Resize的用法

将输入的PIL图像裁剪成指定大小，输入参数为size（int类型或者turple类型），输出为PIL图像

# Resize
transform_resize = transforms.Resize((128, 128))
img_resize = transform_resize(img_PIL)
img_resize = transform_tensor(img_resize)
writer.add_image("Resize", img_resize, 1)
1
2
3
4
5

6. Compose的用法

该方法将所有方法集合到一起，按照顺序依次执行，用法示例如下：

# Compose
transform_compose = transforms.Compose([transform_resize, transform_tensor])
img_resize_2 = transform_compose(img_PIL)
writer.add_image("Compose", img_resize_2, 1)
writer.close()
1
2
3
4
5

7. 所有代码

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

img_path = "C:\\Users\\hp\\PycharmProjects\pythonProject\\Pytorch_Learning\\flower_data\\train\daisy\\5547758_eea9edfd54_n.jpg"

writer = SummaryWriter("logs")
img_PIL = Image.open(img_path)

# ToTensor
transform_tensor = transforms.ToTensor()
img_tensor = transform_tensor(img_PIL)

# Normalize
transform_normal = transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
img_normal = transform_normal(img_tensor)
print(type(img_normal))
writer.add_image("Normalize", img_normal, 1)

# RandomCrop
transform_randomCrop = transforms.RandomCrop((32, 32))
for i in range(0, 10):
    img_randomCrop = transform_randomCrop(img_PIL)
    # print(type(img_randomCrop))
    img_randomCrop = transform_tensor(img_randomCrop)
    writer.add_image("RandomCrop", img_randomCrop, i)

# Resize
transform_resize = transforms.Resize((128, 128))
img_resize = transform_resize(img_PIL)
img_resize = transform_tensor(img_resize)
writer.add_image("Resize", img_resize, 1)

# Compose
transform_compose = transforms.Compose([transform_resize, transform_tensor])
img_resize_2 = transform_compose(img_PIL)
writer.add_image("Compose", img_resize_2, 1)
writer.close()
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38