Pytorch——torch.nn.functional.interpolate函數

最近寫pytorch的時候用到了這個函數：torch.nn.functional.interpolate

一時沒太懂這個函數是幹嘛的，所以看了下pytorch的官方文檔：

torch.nn.functional.interpolate(input, size=None, scale_factor=None, mode='nearest', align_corners=None)：

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

The algorithm used for interpolation is determined by mode.

Currently temporal, spatial and volumetric sampling are supported, i.e. expected inputs are 3-D, 4-D or 5-D in shape.

The input dimensions are interpreted in the form: mini-batch x channels x [optional depth] x [optional height] x width.

The modes available for resizing are: nearest, linear (3D-only), bilinear, bicubic (4D-only), trilinear (5D-only), area

大意就是這個函數是用來上採樣或下采樣，可以給定size或者scale_factor來進行上下采樣。同時支持3D、4D、5D的張量輸入。

插值算法可選，最近鄰、線性、雙線性等等。

來看看這個函數的參數：

• input (Tensor) – the input tensor

• size (int or Tuple[int] or Tuple[int, int] or Tuple[int, int, int]) – output spatial size.

• scale_factor (float or Tuple[float]) – multiplier for spatial size. Has to match input size if it is a tuple.

• mode (str) – algorithm used for upsampling: 'nearest' | 'linear' | 'bilinear' | 'bicubic' |'trilinear' | 'area'. Default: 'nearest'

• align_corners (bool, optional) – Geometrically, we consider the pixels of the input and output as squares rather than points. If set to True, the input and output tensors are aligned by the center points of their corner pixels, preserving the values at the corner pixels. If set to False, the input and output tensors are aligned by the corner points of their corner pixels, and the interpolation uses edge value padding for out-of-boundary values, making this operation independent of input size when scale_factor is kept the same. This only has an effect when mode is 'linear', 'bilinear', 'bicubic' or 'trilinear'. Default: False

舉例語法：

x = nn.functional.interpolate(x, scale_factor=8, mode='bilinear', align_corners=False)