What kind of padding does pytorch use for torch.nn.functional.interpolation, and how to control it?
For example, how to set reflectace padding or zero-padding?
I don’t think that F.interpolate
uses any padding, as it should create the indices corresponding to the desired output shape inside the valid input size.
If you want to pad the input manually, you could use F.pad
, where mode
indicates the desired padding mode.