Common methods in transforms

The sample code is as follows ：

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

writer = SummaryWriter("logs")
img = Image.open(r"D:\pycharm_professional\PycharmProjects\learning_pytorch\images\baby.jpg")


# ToTensor
#  Generate class objects 
trans_totensor = transforms.ToTensor()
#  Pass parameters into the class object and return results 
img_tensor = trans_totensor(img)
#  Add results to events 
writer.add_image("ToTenser",img_tensor)


# Normalize
print(img_tensor[0][0][0])
#  Why is it incoming [0.5,0.5,0.5], Because it has three channels for test pictures 
#  Generate class objects 
trans_norm = transforms.Normalize([0.5,0.5,0.5],[0.5,0.5,0.5])
#  Pass parameters into the class object and return results 
img_norm = trans_norm(img_tensor)
print(img_norm[0][0][0])
#  Add results to events 
writer.add_image("Normalize",img_norm)


# Resize
print(img.size)
#  Generate class objects 
trans_resize = transforms.Resize((512,512))
# img PIL -> resize -> img_resize PIL
img_resize = trans_resize(img)
print(img_resize) # <PIL.Image.Image image mode=RGB size=512x512 at 0x13A7271BE50>
# resize The next image is PIL.Image Data type of , Want to put the changed picture in tensorboard If it shows , Also convert the image to tensor Data type of 
# img_resize PIL -> totensor -> img_resize tensor
img_resize = trans_totensor(img_resize)
writer.add_image("Resize",img_resize)


# Compose - resize -2
trans_resize_2 = transforms.Resize(512)
# PIL -> PIl -> tensor
trans_compose = transforms.Compose([trans_resize_2,trans_totensor])
img_resize_2 = trans_compose(img)
writer.add_image("Resize",img_resize_2,1)


# RandomCrop
trans_random = transforms.RandomCrop(512)
trans_compose_2 = transforms.Compose([trans_random,trans_totensor])
for i in range(10):
    img_crop = trans_compose_2(img)
    writer.add_image("RandomCrop",img_crop,i)


writer.close()

The above code results in tensorboard visualization ：

tips

• In fact, you can find that ,ToTensor、Normalize、Resize、Compose The method of using is to create a corresponding class object , Then pass the picture into the class object , Return a result .

#  Example ：
#  Generate class objects  
trans_totensor = transforms.ToTensor()
#  Pass parameters into the class object and return results 
img_tensor = trans_totensor(img)
#  Add results to events 
writer.add_image("ToTenser",img_tensor)

• One thing to note is that , Be sure to find out before using a method , What parameter types are required for this method ！ It is important to note what type of input a method has , What type of output ！

#  Example :
# Resize
print(img.size)
#  Generate class objects 
trans_resize = transforms.Resize((512,512))
# img PIL -> resize -> img_resize PIL
img_resize = trans_resize(img)
print(img_resize) # <PIL.Image.Image image mode=RGB size=512x512 at 0x13A7271BE50>
# resize The next image is PIL.Image Data type of , Want to put the changed picture in tensorboard If it shows , Also convert the image to tensor Data type of 
# img_resize PIL -> totensor -> img_resize tensor
img_resize = trans_totensor(img_resize)
writer.add_image("Resize",img_resize)

Like Resize It's about putting PIL Carry out size transformation , After the size change PIL Image processing ToTensor Transformation , Show it again , Not directly to tensor Type of image data Resize.

• Remember that in pycharm When writing code in, you must master ctrl + alt and ctrl + p Use , These two shortcuts can reduce a lot of problems
• There is also the use of a less commonly used method , The first thing you can think of is to jump back to the source code , Look at the official explanation .

summary ：
1. Focus on input and output types
2. Read more official documents
3. When you don't know the return value ：

• print()
• print(type())
• debug