I have a question about F.interpolate, same as torch.nn.functional.interpolate.
when we are performing downsampling using F.interpolate(tensor, size, mode=‘bilinear’, align_corners=False), how does it working?
Is it performing average pooling or max pooling?
And is anti-aliasing necessary? aliasing will be occurred?
Additionally, what’s the method for applying anti-aliasing? Is Low Pass Filtering with Gaussian kernel?