Fabian Isensee

You need to check your system. Is the CPU fully loaded? Are you running into I/O issues? What do the plans look like? How many input channels does your model...

My suggestions would be to not change things without good reason. Num threads must be se correctly for the CPU to be able to prepare training data fast enough. If...

It seems like your training dataset was not properly created. Have you followed the instructions closely? You might also want to take a look at the existing datasets (MSD for...

This would be rather difficult to explain in detail. To give you food for thought: imaging you have label 1 and label 2 and they overlap, the solution would be...

Hey it seems like the dataset is either really really small OR has conflicting labels (essentially bad ground truth segmentations). I would check the latter first!

Knowing prostate tumor segmentationa bit I wouldn't be surprised if your ground truth labels are inaccurate. Registrating multiple modalities is a real pain here. @Kobalt93 can maybe help!?

Simply convert the labels to [0, 1, 2, 3, 4, ...]?

Yeah just open the segmentations with any tool, modify the labels and then save again. That's not hard. Example: ``` seg = sitk.ReadImage(FILE) seg_npy = sitk.GetArrayFromImage(seg) seg_converted = np.zeros_like(seg_npy) seg_converted[seg_npy...

24 represents the patch size in the first dimension. If your first dimension consists of layers then yes that's the layers. If the number of slices/layers is less than 24...

During preprocessing each image is processed as one, it is not divided into patches. During training, an infinite number of random (!) patches is cropped from the preprocessed image. Patches...