（三）Transforms的使用

1.transforms的结构与用法

2.transforms的代码实战

2.1 ToTensor代码

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

img_path = "/Users/songyuxi/PycharmProjects/pythonProject/hymenoptera_data/train/ants/0013035.jpg"
img_pil = Image.open(img_path)  # jpeg类型
transform = transforms.ToTensor()  # 声明ToTensor类的对象
img_tensor = transform(img_pil)  # 将jpeg类型转换为tensor类型
writer = SummaryWriter(log_dir="logs")
writer.add_image(tag="tensor_img", img_tensor=img_tensor, global_step=2)
writer.close()

2.2Normalize代码

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

img_path = "/Users/songyuxi/PycharmProjects/pythonProject/images/01.jpeg"
img_pil = Image.open(img_path)  # jpeg类型

# totensor
transform = transforms.ToTensor()  # 声明ToTensor类的对象
img_tensor = transform(img_pil)  # 将jpeg类型转换为tensor类型

# normalize
norm = transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])  # 声明Normalize类对象
img_tensor_norm = norm(img_tensor)  # 将tensor类型图像归一化

writer = SummaryWriter(log_dir="logs")
writer.add_image(tag="tensor_img", img_tensor=img_tensor, global_step=1)
writer.add_image(tag="tensor_img_norm", img_tensor=img_tensor_norm, global_step=1)
writer.close()

2.3Resize的代码

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

img_path = "/Users/songyuxi/PycharmProjects/pythonProject/images/01.jpeg"
img_pil = Image.open(img_path)  # jpeg类型

# totensor
transform = transforms.ToTensor()  # 声明ToTensor类的对象
img_tensor = transform(img_pil)  # 将jpeg类型转换为tensor类型

# normalize
norm = transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])  # 声明Normalize类对象
img_tensor_norm = norm(img_tensor)  # 将tensor类型图像归一化

# resize
resize = transforms.Resize((512, 512))
img = resize(img_pil)  # 将img_pil由高640，宽368 转换为高512，宽512
print(img_pil.size)
print(img.size)
img_tensor_resize = transform(img)  # 将jpeg类型转换为tensor类型

writer = SummaryWriter(log_dir="logs")
writer.add_image(tag="tensor_img", img_tensor=img_tensor, global_step=1)
writer.add_image(tag="tensor_img_norm", img_tensor=img_tensor_norm, global_step=1)
writer.add_image(tag="tensor_img_resize", img_tensor=img_tensor_resize, global_step=1)
writer.close()

2.4Compose的代码

Compose（）中的参数需要是一个列表，Python中列表的表示形式为[数据1，数据2，数据3，…]。在Compose中，数据需要是transforms类型，所以得到Compose（[transforms参数1，transforms参数2，…]）

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

img_path = "/Users/songyuxi/PycharmProjects/pythonProject/images/01.jpeg"
img_pil = Image.open(img_path)  # jpeg类型

# totensor
transform = transforms.ToTensor()  # 声明ToTensor类的对象
img_tensor = transform(img_pil)  # 将jpeg类型转换为tensor类型

# normalize
norm = transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])  # 声明Normalize类对象
img_tensor_norm = norm(img_tensor)  # 将tensor类型图像归一化

# resize
resize = transforms.Resize((512, 512))
img = resize(img_pil)  # 将img_pil由高640，宽368 转换为高512，宽512
print(img_pil.size)
print(img.size)
img_tensor_resize = transform(img)  # 将jpeg类型转换为tensor类型

# compose  (640,368)->(512,512)->tensor->normalize
compose = transforms.Compose([resize, transform, norm])
img_compose = compose(img_pil)

writer = SummaryWriter(log_dir="logs")
writer.add_image(tag="tensor_img", img_tensor=img_tensor, global_step=1)
writer.add_image(tag="tensor_img_norm", img_tensor=img_tensor_norm, global_step=1)
writer.add_image(tag="tensor_img_resize", img_tensor=img_tensor_resize, global_step=1)
writer.add_image(tag="tensor_img_compose", img_tensor=img_compose, global_step=1)
writer.close()

2.5RandomCrop的代码

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

img_path = "/Users/songyuxi/PycharmProjects/pythonProject/images/01.jpeg"
img_pil = Image.open(img_path)  # jpeg类型

# totensor
transform = transforms.ToTensor()  # 声明ToTensor类的对象
img_tensor = transform(img_pil)  # 将jpeg类型转换为tensor类型

# normalize
norm = transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])  # 声明Normalize类对象
img_tensor_norm = norm(img_tensor)  # 将tensor类型图像归一化

# resize
resize = transforms.Resize((512, 512))
img = resize(img_pil)  # 将img_pil由高640，宽368 转换为高512，宽512
print(img_pil.size)
print(img.size)
img_tensor_resize = transform(img)  # 将jpeg类型转换为tensor类型

# compose  (640,368)->(512,512)->tensor->normalize
compose = transforms.Compose([resize, transform, norm])
img_compose = compose(img_pil)

# RandomCrop
randomcrop = transforms.RandomCrop(300)
compose1 = transforms.Compose([randomcrop, transform])
img_random = compose1(img_pil)

writer = SummaryWriter(log_dir="logs")
writer.add_image(tag="tensor_img", img_tensor=img_tensor, global_step=1)
writer.add_image(tag="tensor_img_norm", img_tensor=img_tensor_norm, global_step=1)
writer.add_image(tag="tensor_img_resize", img_tensor=img_tensor_resize, global_step=1)
writer.add_image(tag="tensor_img_compose", img_tensor=img_compose, global_step=1)
writer.add_image(tag="tensor_img_random", img_tensor=img_random, global_step=1)
writer.close()

3.transforms包下的常用类

3.1 Compose

3.2 ToTensor

3.3 Normalize

3.4 Resize

• CenterCrop