Pytorch framework learning record 3 - the use of Transform

Pytorch框架学习记录3——Transform的使用

1. 导入

from torchvision import transforms

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()

3. Normalize的用法

用均值和标准差归一化张量图像,参数：mean, std;The input image type is 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)

4. RandomCrop 的用法

随机裁剪,参数sizeIndicates the size of the randomly cropped image,The input image type is PIL类型,The type of the output is alsoPIL类型.

The example below is a random crop of an image10次,The size of the image after each crop is 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)

5. Resize的用法

将输入的PILCrop the image to the specified size,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)

6. Compose的用法

This method brings all the methods together,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()

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()