Hello, I wanted to ask regarding the border parameter for inference. If I understood it is necessary to specify a minimum when the output patch size is less than the input, in order to compensate the 'void regions' that would appear between output window patches if we were to set border to 0. But this being a minimum means we can specify bigger values, which in practice I believe means that the output patch windows will overlap in a certain degree (this should also happen if we set a border>0 when input and output sizes are the same). I read this answer saying that this had an effect of smoothing the result (https://github.com/NifTK/NiftyNet/issues/103). My question is to understand what is exactly happening when we do this, my initial guess is that the non-overlaping regions remain the same but the overlaping regions are averaged in some way, and hence the smoothing. Still I´m not sure about if anything of what I wrote is the actual thing, these are my thoughts on how it might be working, I would like to have the correct picture here for correct testing.
Hello, I wanted to ask regarding the border parameter for inference. If I understood it is necessary to specify a minimum when the output patch size is less than the input, in order to compensate the 'void regions' that would appear between output window patches if we were to set border to 0. But this being a minimum means we can specify bigger values, which in practice I believe means that the output patch windows will overlap in a certain degree (this should also happen if we set a border>0 when input and output sizes are the same). I read this answer saying that this had an effect of smoothing the result (https://github.com/NifTK/NiftyNet/issues/103). My question is to understand what is exactly happening when we do this, my initial guess is that the non-overlaping regions remain the same but the overlaping regions are averaged in some way, and hence the smoothing. Still I´m not sure about if anything of what I wrote is the actual thing, these are my thoughts on how it might be working, I would like to have the correct picture here for correct testing.
Thanks in advance.