We found out that the "scale transfer" done by knowledge transfer can harm the detection precision (a little) if the scales of the source and target volumes are very similar. We could allow users to apply "knowledge transfer" without scale transfer, effectively using the training annotations as-is. This could also be used if no scale information is available (i.e. "existing annotations" but from another volume).
We found out that the "scale transfer" done by knowledge transfer can harm the detection precision (a little) if the scales of the source and target volumes are very similar. We could allow users to apply "knowledge transfer" without scale transfer, effectively using the training annotations as-is. This could also be used if no scale information is available (i.e. "existing annotations" but from another volume).