How to change the source code to execute 3d_lowres and 3d_cascade_fullres network

[tea321000]

I want to test the performance of the Brats dataset on 3d_lowres and 3d_cascade_fullres network, and when I execute these commands directly, I will be prompted that they do not need to be used. I changed self.how_much_of_a_patient_must_the_network_see_at_stage0 = 4 to self.how_much_of_a_patient_must_the_network_see_at_stage0 = 1 in experiment_planner_baseline_3DUNet.py, but still failed to pass. How could I execute 3d_lowres and 3d_cascade_fullres？

[FabianIsensee]

Hi, it makes no sense to run the cascade on BraTS. This will not improve results. If you really want to run the cascade you probably also need to chanke these lines

            if 2 * np.prod(new['median_patient_size_in_voxels'], dtype=np.int64) < np.prod(
                    self.plans_per_stage[0]['median_patient_size_in_voxels'], dtype=np.int64):

in ExperimentPlanner. Best, Fabian

[FabianIsensee]

Probably

            if True:

will do

[tea321000]

Thanks for your suggestion, I just want to test the performance of the network using the available dataset, although that may not be helpful. I've finished training the 3d_lowres network, but when I run nnUNet_train 3d_cascade_fullres nnUNetTrainerV2CascadeFullRes TaskXXX all, it prompts AssertionError: seg from prev stage missing: /home/user062/nnunet/dataset/nnUNet_trained_models/nnUNet/3d_lowres/Task100_BraTS2020/nnUNetTrainerV2__nnUNetPlansv2.1/pred_next_stage/BraTS20_Training_001_segFromPrevStage.npz.

Do you know how to fix it? Do I have to run

nnUNet_train 3d_lowres nnUNetTrainerV2 TaskXXX 0
nnUNet_train 3d_lowres nnUNetTrainerV2 TaskXXX 1
nnUNet_train 3d_lowres nnUNetTrainerV2 TaskXXX 2
nnUNet_train 3d_lowres nnUNetTrainerV2 TaskXXX 3
nnUNet_train 3d_lowres nnUNetTrainerV2 TaskXXX 4

separately instead of nnUNet_train 3d_lowres nnUNetTrainerV2 TaskXXX all?

[FabianIsensee]

Yes you need to run the 5 folds separately. Best, Fabian

[tea321000]

Okay, thanks for your reply!