I was reading through the nnU-net paper, it is mentioned that the resampling along the out-of-plane axis is not required for slice-based processing. The target spacing for processing is determined based on the median spacing of training cases for each axis before deciding on the architecture and training process and the resampled is already done based on that. I have noticed that the image shape along z direction is not the same after resampling and I am planning to use 2D U-net for training, so how the model would use those data (not resampled along z-direction) if the pre-processed data already resampled on out of plane axis?
Hi Fabian,
I was reading through the nnU-net paper, it is mentioned that the resampling along the out-of-plane axis is not required for slice-based processing. The target spacing for processing is determined based on the median spacing of training cases for each axis before deciding on the architecture and training process and the resampled is already done based on that. I have noticed that the image shape along z direction is not the same after resampling and I am planning to use 2D U-net for training, so how the model would use those data (not resampled along z-direction) if the pre-processed data already resampled on out of plane axis?
Please see the attached image.