Tensor size error in MedNextv1 trainer

[mariaperezf]

Hello, I am having trouble with the following Runtime Error:

RuntimeError: The size of tensor a (3) must match the size of tensor b (4) at non-singleton dimension 2

When trying to train a model using MedNext with the command:

mednextv1_train 3d_fullres nnUNetTrainerV2_MedNeXt_S_kernel3 999 0

[Helen52520]

Got the same error. Did you fix the problem? Thanks!

[mariaperezf]

Got the same error. Did you fix the problem? Thanks!

No, not really. Although it does work if you use the preprocessing from the MICCAI 2023 that is shown in the READEME, so I guess it has something to do with the spacing:

mednextv1_plan_and_preprocess -t YOUR_TASK -pl3d ExperimentPlanner3D_v21_customTargetSpacing_1x1x1 -pl2d ExperimentPlanner2D_v21_customTargetSpacing_1x1x1

mednextv1_train 3d_fullres TRAINER TASK_NUMBER FOLD -p nnUNetPlansv2.1_trgSp_1x1x1

[Helen52520]

Got the same error. Did you fix the problem? Thanks!

No, not really. Although it does work if you use the preprocessing from the MICCAI 2023 that is shown in the READEME, so I guess it has something to do with the spacing:

mednextv1_plan_and_preprocess -t YOUR_TASK -pl3d ExperimentPlanner3D_v21_customTargetSpacing_1x1x1 -pl2d ExperimentPlanner2D_v21_customTargetSpacing_1x1x1

mednextv1_train 3d_fullres TRAINER TASK_NUMBER FOLD -p nnUNetPlansv2.1_trgSp_1x1x1

I successfully got it running with your command, thank you! By the way, may I ask if you're also working with multi-modality? The original command (-p nnUNetPlansv2.1) works for single-channel, but it fails for multi-modality.