MIC-DKFZ
nnunetv2's source dataset and target dataset when pretraining
Hello, thank you for your work for nnunetv2.
Firstly, i will introduce two datasets: the first one is source dataset which has 7 classes, the second one is target dataset which has 2 classes.
when i run this command: nnUNetv2_train SOURCE_DATSET CONFIG all -p TARGET_PLANS_IDENTIFIER,I found in the pseudo dice there are 7 classes but the last 5 classes are nan lick this[0.8643, 0.322, nan, nan, nan, nan, nan] . I think the reason is that I am training on the target dataset which only has 2 classes. After some study, I think the problem is the source_dataset and target_dataset.
For example: nnUNetv2_move_plans_between_datasets -s TARGET_DATASET -t SOURCE_DATASET -sp TARGET_PLANS_IDENTIFIER -tp SOURCE_PLANS_IDENTIFIER, in this command, I think -s means source so it will stand for the target_dataset and -t means target so it will stands for the source_dataset but the readme file gives me a different suggestion.
In summary, I have 2 questions: 1:the pseudo dice is [0.8643, 0.322, nan, nan, nan, nan, nan], whether this means i am training on a dataset having 2 classes? 2:If the first one is yes. What is the wrong? And should I exchange the target dataset with source dataset?
