【pytorch】transforms. Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))

ransform.Normalize(): The tensor image is normalized by mean and standard deviation

Often see

transforms.Compose([transforms.ToTensor(), 
                    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

that transform.Normalize() How does it work ？ Take the above code for example ,

---

ToTensor() Did two things ：

• Change the gray scale from 0-255 Change to 0-1 Between , It normalizes each value to [0,1], The normalization method is relatively simple , Divide directly by 255 that will do
• take shape by (H,W, C) Of nump.ndarray or img To shape by (C, H, W) Of tensor

---

transforms.Normalize()
transforms.Normalize(std=(0.5,0.5,0.5),mean=(0.5,0.5,0.5)), Its function is to normalize the input to (0,1), Then use the formula "(x-mean)/std", Distribute each element to (-1,1)

image=(image-mean)/std

among mean and std Pass respectively (0.5,0.5,0.5) and (0.5,0.5,0.5) Make a designation . The original 0-1 minimum value 0 Has become a (0-0.5)/0.5=-1, And the maximum 1 Has become a (1-0.5)/0.5=1.

But I think many codes are like this ：
torchvision.transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
How does this group of values come from ？ This set of values is from imagenet It's calculated by sampling in the training center .

summary ：

Go over it normalize() After the transformation, the mean value becomes 0 The variance of 1（ In fact, the maximum and minimum values are 1 and -1）

Each sample image becomes a mean value of 0 The variance of 1 The standard normal distribution of , This is the most common （ The greatest value of scientific research ） Sample data of