The pytorch doc says that “grid specifies the sampling pixel locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]”. But why do they let grid_sample function specify the normalized pixel locations?
Thank you very much for any help!