Fabian Isensee comments

Results 453 comments of


                                            Fabian Isensee

Epoch time doubles when running separate tasks on different GPUs

Also read and run this: https://github.com/MIC-DKFZ/nnUNet/blob/a4af4c05be482b764131d38fae5308ace63eb39f/documentation/expected_epoch_times.md#results

Unable to crop

Indeed, thanks @NB-UCLouvain ! Cropping is run at all times, but sometimes it just doesn't do anything and that's OK.

Error using nnUNetTrainerV2_Loss_Dice, nnUNetTrainerV2_focalLoss, and nnUNetTrainerV2_Adam_lr_3en4

I think your python version is too old. Try 3.8 or 3.9

Patch size NIH (TCIA Pancreas) data-set bigger than 128^3

Hey, do we really say the maximum patch size is 128^3? That would be incorrect. Is that in the nature methods paper? The patch size is configures based on the...

Obtain probabilities mask

There is no threshold since we use argmax (!). So in a 2-class case one could argue that the threshold is 0.5. In multi-class segmentation there is no clear threshold...

Assertion Error: If does not seem like you trained all the folds! --> But trained all folds

Did you maybe use a different trainer? If so then you need to specify which one. Can you also share the path of your trained model?

Loading pre-computed distance transform during training

Have you seen this branch? https://github.com/MIC-DKFZ/nnUNet/tree/distance_maps It is horribly outdated but it does exactly that. This branch was initially made for @JunMa11, I am not sure whether he is sill...

importlib.metadata.PackageNotFoundError: nnunet

Can you please start from a fresh virtual environment? Conda is preferred. Does the error still occur?

Add progress bar to the download of the pre-trained weights (working solution inside)

thanks for the suggestion. I added it to the master

Add progress bar to the download of the pre-trained weights (working solution inside)

It worked for me. Does it not work for you?