In torch.nn.functional.interpolate what's the difference between the modes linear and bilinear?
To me, these are usually synonyms with regards to image resizing...
Pytorch is explicitly differentiating between 1d interpolation (linear) and 2d interpolation (bilinear).
They differ in the the dimensionality of the input argument they are allowed to work on ( see here ). Specifically, linear works on 3D inputs and bilinear works on 4D inputs because the first two dimensions (mini-batch x channels) are understood not to be interpolated.