Pytorch: 插值和上采样

nn.Upsample

定义

import torch.nn as nn
import torch.nn.functional as F

nn.Upsample(size=None, scale_factor=None, mode='nearest', align_corners=None)
F.upsample(input, size=None, scale_factor=None, mode='nearest', align_corners=None)

参数

align_corners

上图是source pixel为44，上采样为target pixel为88的对齐和不对齐两种情况。当设置align_corners = True时，会对齐左上角元素，即输入的左上角元素是一定等于输出的左上角元素。

形状
或者指定输出的size(W_out)、size(H_out, W_out)、size(D_out, H_out, W_out)

nn.functional.interpolate

定义
Down/up samples the input to either the given size or the given scale_factor

import torch.nn.functional as F
F.interpolate(input, size=None, scale_factor=None, mode='nearest', align_corners=None, recompute_scale_factor=None)

参数

形状
或者指定输出的size(W_out)、size(H_out, W_out)、size(D_out, H_out, W_out)

Warning

warnings.warn(“nn.functional.upsample is deprecated(不赞成，反对). Use nn.functional.interpolate instead.”)
现在都建议使用interpolate方法实现上采样，所以尽量用interpolate