Does resize affect image details?

Im working on image filtering right now and wanted to know if resizing in anyway has any kind of filtering effect on the images?

Resizing an image can be seen as a subsampling operation, which could need an anti-alasing filter to prevent aliasing effects (if the image frequencies start to overlap due to the sampling).
This Wikipedia article gives you a high-level overview for a couple of interpolation techniques.

I haven’t seen a lot of thought in the machine learning area regarding aliasing effects in images and would guess that for a lot of use cases the default of image processing libraries seems to work. :wink:

Hi @ptrblck, does torch.nn.functional.interpolate with mode='bicubic' or 'bilinear' have any anti-aliasing built-in? Thank you.

I don’t think an anti-aliasing filter is used and just the resampling is performed based on this implementation.