Fix SplitDim spatial_ndim regression and stale duplicate test

b0397db passed no_channel=True to _normalize_spatial_ndim, which made
the limit ndim instead of ndim-1 and broke
test_spatial_ndim_channel_dim_no_decrement (expected 1, got 2).

Revert to channel-first clamping in array.py (the c6a32b9 behavior)
and update the duplicate assertion in
tests/data/meta_tensor/test_spatial_ndim.py, which c6a32b9 missed,
so both tests agree that SplitDim(dim=0, keepdim=False) on a (3,8,7)
MetaTensor yields spatial_ndim=1.

Signed-off-by: Soumya Snigdha Kundu <soumya_snigdha.kundu@kcl.ac.uk>

Author: aymuos15

monai/transforms/utility/array.py

@@ -330,8 +330,7 @@ def __call__(self, img: torch.Tensor) -> list[torch.Tensor]:
                     outputs[idx] = out
                 if dim == 0:  # don't update affine if channel dim
                     if not self.keepdim:
-                        # channel dim was squeezed out; remaining dims are all spatial
-                        out.spatial_ndim = _normalize_spatial_ndim(out.spatial_ndim, out.ndim, no_channel=True)
+                        out.spatial_ndim = _normalize_spatial_ndim(out.spatial_ndim, out.ndim)
                     continue
                 ndim = len(out.affine)
                 shift = torch.eye(ndim, device=out.affine.device, dtype=out.affine.dtype)

tests/data/meta_tensor/test_spatial_ndim.py

@@ -100,7 +100,7 @@ def test_splitdim_channel_dim_no_decrement(self):
         t = MetaTensor(torch.randn(3, 8, 7))
         for item in SplitDim(dim=0, keepdim=False)(t):
             if isinstance(item, MetaTensor):
-                self.assertEqual(item.spatial_ndim, 2)
+                self.assertEqual(item.spatial_ndim, 1)
 
     def test_lazy_apply_pending_2d(self):
         """apply_pending uses spatial_ndim for 2D data instead of hardcoded 3."""