[Question] Label files with multiple channels

[y0m4r0ll]

❓ Question: label files with multiple channels

Hi!

I have label nifti files with multiple channels (each channel is one instance as the segmentations overlap in some areas), but nnDetection fails at the prep step:
Log:
2022-08-18 15:31:00.090 | INFO | nndet.preprocessing.crop:process_data:223 - Processing case case_0010
2022-08-18 15:31:05.107 | INFO | nndet.preprocessing.crop:crop:288 - Shape before crop (1, 350, 450, 450); after crop (1, 262, 375, 375); spacing [1. 1. 1.]
2022-08-18 15:31:07.012 | INFO | nndet.preprocessing.crop:process_data:237 - exception in: case_0024: operands could not be broadcast together with shapes (1,262,375,375,2) (1,262,375,375)
(I have tried shape (1,2,262,375,375) and it produced the same error.)

The masks were generated with SimpleITK from numpy arrays.
Is there something I am missing or is this just not supported? Thanks in advance!

[mibaumgartner]

Dear @y0m4r0ll ,

overlapping object in multi channel segmentations files are currently not supported in nndetection.

Best,
Michael