Hello there,
Thanks for the great work!
I noticed that the dataset provides cropped and 1mm resampled images. However, it is common knowledge that iterative solvers benefit greatly from higher resolution and more context, something that deep learning methods struggle with due to non-standard voxel sizes and image shapes. We have observed this internally with a few tools as well.
Can the data also be released in a "raw" format without any cropping or downsampling? This will allow baselines that can work with high resolution data to perform better.
Hello there,
Thanks for the great work!
I noticed that the dataset provides cropped and 1mm resampled images. However, it is common knowledge that iterative solvers benefit greatly from higher resolution and more context, something that deep learning methods struggle with due to non-standard voxel sizes and image shapes. We have observed this internally with a few tools as well.
Can the data also be released in a "raw" format without any cropping or downsampling? This will allow baselines that can work with high resolution data to perform better.