Poor segmentation results with TotalSegmentator.

[windygoo]

import os
import SimpleITK as sitk
import numpy as np

data = np.load("ct/trn00022.npz")
ct_data = data["ct"]

ct = sitk.GetImageFromArray(ct_data)
ct = sitk.Cast(ct, sitk.sitkInt16)

ct.SetSpacing((0.8, 0.8, 0.8))

orientation = [1, 0, 0, 0, 1, 0]
row_cosine, col_cosine = orientation[:3], orientation[3:6]
z_cosine = np.cross(row_cosine, col_cosine).tolist()
ct.SetDirection(row_cosine + col_cosine + z_cosine)

sitk.WriteImage(ct, "radchest_debug.nii.gz")

I use the above code to convert the data format from npz to nii.gz.
Then, I use TotalSegmentator to predict the masks of prepared nii.gz images. However, I find the segmentation results are poor. How can I solve it? Thanks!