一、常见的transform的使用

1. python中call的用法

image.png

tips：pycharm中使用ctrl + p 可以查看函数所需参数

二、 Totensor和Normalize的使用

  1. 读取图片 ```python from PIL import Image from torchvision import transforms from tensorboardX import SummaryWriter

writer = SummaryWriter(“logs”) path = r”test.png”

四通道图片时使用convert方法

img = Image.open(path).convert(“RGB”)

  2. 2. 使用transform
  3. ```python
  4. # ToTensor
  5. trans_totensor = transforms.ToTensor()
  6. img_tensor = trans_totensor(img)
  7. writer.add_image("ToTensor", img_tensor)
  9. # Normalize
  10. trans_norm = transforms.Normalize([0.5,0.5,0.5], [0.5,0.5,0.5])
  11. img_norm = trans_norm(img_tensor)
  12. writer.add_image("Normalize", img_norm)
  14. # Resize
  15. trans_resize = transforms.Resize((512, 512))
  16. img_resize = trans_resize(img)
  17. img_totensor2 = trans_totensor(img_resize)
  18. writer.add_image("Resize", img_totensor2)
  20. # Resize 使用Compose方法
  21. trans_compose = transforms.Compose([trans_resize, trans_totensor])
  22. img_resize = trans_compose(img)
  23. writer.add_image("Resize-Compose",img_resize, 1)
  25. writer.close()

tips：如果标准化时报错：
image.png
说明图片为四通道图片，则读取时需要使用 Image.open(path).convert("RGB")
具体见博客：戳！

jupyter notebook 执行终端命令：!tensorboard --logdir=logs --port=6007