[pytorch] simple use of interpolate

interpolate It is used for interpolation , Common use is for up sampling (upsampling); Of course, you can also take samples (downsampling)

The official introduction is as follows

The parameter list is as follows

Easy to use

The following is a brief introduction to use

size and scale_factor You only need to provide one of the two parameters ;

The difference between them is ,size Specifies the interpolated shape; and scale_factor Just provide a coefficient to multiply or subtract ;

mode Mainly refers to the interpolation algorithm , The default is "nearest", It is similar to repeat operation , That is, copy the latest element to realize interpolation , But it must be applied to the input tensor is 4 Dimensional ;
Another common value is “linear”, It does linear interpolation

The following is a simple example

Examples 1

scale_factor Is to specify the expansion factor of the last two dimensions , As below scale_factor=(2,2), It's an input tensor shape The last two dimensions of , Times, respectively, 2;

import torch
from torch.nn import functional as F

x=torch.randint(1,10, (1, 3,2,2)).type(torch.float)
print(x)
x = F.interpolate(x, scale_factor=(2,2), mode='nearest')
print(x, x.shape)

Examples 2

If scale_factor Whether it is OK to enter only one dimension ？ Tolerable .
The following example scale_factor=(2), Its effect is scale_factor=(2,2)

import torch
from torch.nn import functional as F

x=torch.randint(1,10, (1, 3,2,2)).type(torch.float)
print(x)
x = F.interpolate(x, scale_factor=(2), mode='nearest') 
print(x, x.shape)

Examples 3

Don't use scale_factor, While using size;
In the following example , adopt size=(4,4), Specifies the final two-dimensional target size of the input tensor ; In fact, in this example , Its function is the same as the above example .

import torch
from torch.nn import functional as F

x=torch.randint(1,10, (1, 3,2,2)).type(torch.float)
print(x)

x = F.interpolate(x,size=(4,4), mode='nearest') print(x, x.shape)

Examples 4

Use mode=‘linear’; Note that the input tensor needs to be 3 dimension , Then interpolation only works on the last dimension

import torch
from torch.nn import functional as F

x=torch.randint(1,10, (3,2,2)).type(torch.float)
print(x)
x = F.interpolate(x,size=4, mode='linear') #, scale_factor=(2)
print(x, x.shape)

Examples 5

Use scale_factor, The effect same as above

import torch
from torch.nn import functional as F

x=torch.randint(1,10, (3,2,2)).type(torch.float)
print(x)
x = F.interpolate(x,scale_factor=2, mode='linear') 
print(x, x.shape)