Suppose that I train my image dataset on CNN, but the resolution of the image varies significantly on the dataset.
In this case, should I scale the images up to the image that has the maximum resolution, or scale down to the lowest resolution? In that case should I scale up/down the whole images even if the highest likelihood of the whole samples are somewhere in the middle of the distribution of the resolution?
Or should I use another technique to deal with the varying resolution problem?
