Segmentation Validity Issue in Docker for Pelvic Fracture CT

I encountered an error while trying out the algorithm. It prompted:

"The image produced in 'images/pelvic-fracture-ct-segmentation' is not valid. Image segments could not be determined, ensure the voxel values are integers and that it contains no more than 64 segments. Ensure the image has the minimum and maximum voxel values set as tags if the image is a TIFF file."

However, I checked the MHA file output when running Docker locally. I confirmed that its data type is int16 (consistent with the gold standard), and the number of segments does not exceed 64. The MHA file generated locally can also be read normally using ITK and visualized using ITK-SNAP. I am now unsure where the problem lies.

Here is some information about the segmentation results I output locally:

output_path = ("/home/ypy/Code/PENGWIN-example-algorithm-main/PENGWIN-challenge-packages/preliminary-development-phase-ct/test/output/images/pelvic-fracture-ct-segmentation/071.mha")

loaded_image = sitk.ReadImage(output_path)

loaded_array = sitk.GetArrayFromImage(loaded_image)

loaded_array.shape

(319, 209, 351)

loaded_array.dtype

dtype('int16')

np.unique(loaded_array)

array([ 0, 1, 2, 11, 12, 21], dtype=int16)

loaded_array.min()

0

loaded_array.max()

21

I thought maybe you could set dtype to torch.int8.

i change the dtype to int8 and it finally works

Thank you for the feedback! I will include this info in the guideline.