[pytorch入门] 3. torchvision中的transforms

torchvision中的transforms

是transforms.py工具箱，含有totensor、resize等工具
用于将特定格式的图片转换为想要的图片的结果，即用于图片变换

用法

在transforms中选择一个类创建对象，使用这个对象选择相应方法进行处理

能够选择的类

列表

[
    "Compose",
    "ToTensor",  # 转化为tensor类型
    "PILToTensor",
    "ConvertImageDtype",
    "ToPILImage", # tenor或ndarray转换为PIL image
    "Normalize",  # 归一化（用均值或标准差归一化tensor类型的image）
    "Resize",
    "CenterCrop",
    "Pad",
    "Lambda",
    "RandomApply",
    "RandomChoice",
    "RandomOrder",
    "RandomCrop",
    "RandomHorizontalFlip",
    "RandomVerticalFlip",
    "RandomResizedCrop",
    "FiveCrop",
    "TenCrop",
    "LinearTransformation",
    "ColorJitter",
    "RandomRotation",
    "RandomAffine",
    "Grayscale",
    "RandomGrayscale",
    "RandomPerspective",
    "RandomErasing",
    "GaussianBlur",
    "InterpolationMode",
    "RandomInvert",
    "RandomPosterize",
    "RandomSolarize",
    "RandomAdjustSharpness",
    "RandomAutocontrast",
    "RandomEqualize",
    "ElasticTransform",
]

使用

• NT
  • 关注输入和输出类型
  • 多看官方文档（可以进源码去看）

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

writer = SummaryWriter('logs')
img = Image.open('./dataset2/train/ants_image/5650366_e22b7e1065.jpg')
print(img)

# ToTensor
trans_totensor = transforms.ToTensor()
img_tensor = trans_totensor(img)       # 将PIL对象转换为tensor对象
writer.add_image("tensor_img",img_tensor)

# Normalize 归一化：（输入-均值）/标准差
print(img_tensor[0][0][0])
trans_norm = transforms.Normalize([0.5,0.5,0.5],[0.5,0.5,0.5]) # 要提供一个均值一个标准差
img_norm = trans_norm(img_tensor)
print(img_norm[0][0][0])
writer.add_image("norm_img",img_norm)

# Resize 调整尺寸：(h,w)
print(img.size)
trans_resize = transforms.Resize((512,512))
    # img PIL -> resize -> img_resize PIL
img_resize = trans_resize(img)  # 这里要传入的是PIL image类型
    # img_resize PIL -> totensor -> img_resize tensor
img_resize = trans_totensor(img_resize)
print(img_resize)


# Compose  -  resize 第二种用法： 等比缩放
trans_resize_2 = transforms.Resize(512)
    # PIL -> PIL -> tensor
trans_compose = transforms.Compose([trans_resize_2, trans_totensor])  # 从右往左执行，后面参数的输入和前面参数的输出是一致的
img_resize_2 = trans_compose(img)
writer.add_image("Resize", img_resize_2, 1)

# RandomCrop 随机裁剪
trans_random = transforms.RandomCrop(256) # 传一个int：代表正方形，传一个tuple：代表矩形
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()

进入调试，就会发现tensor_img中有一些反向传播的钩子：

以及使用的设备：

可以理解为tensor类型包含了神经网络训练所需要的一些参数

其他

类中的__call__方法:对象可以直接用括号时自动用的该方法（区别于普通方法需要用“.”调用）

class Person:
    def __call__(self, name):
        print("__call__:", name)
    
    def hello(self, name):
        print("hello",name)

p = Person()
p("jerry")
p.hello('tom')