When transforming DataTree elements, discard scales resulting in zero shape (#948)

* Add test showing that DataTree transform fails when transformed scales have zero shape

* Handle resulting scales of shape zero in DataTree transform and make associated test pass

* Make test name more specific

* Update tests/core/operations/test_transform.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Update src/spatialdata/_core/operations/transform.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* tiny improvements

---------

Co-authored-by: Luca Marconato <m.lucalmer@gmail.com>
Co-authored-by: LucaMarconato <2664412+LucaMarconato@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: 3 people

src/spatialdata/_core/operations/transform.py

@@ -381,6 +381,18 @@ def _(
         transformed_dask, raster_translation_single_scale = _transform_raster(
             data=xdata.data, axes=xdata.dims, transformation=composed, **kwargs
         )
+
+        # if a scale in the transformed data has zero shape, we skip it
+        if 0 in transformed_dask.shape:
+            if k == "scale0":
+                raise ValueError(
+                    "The transformation leads to zero shaped data even at the highest resolution level. "
+                    "Check the scaling component of the transformation."
+                )
+            # no risk of skipping a scale (e.g. scale1) but not the next ones (e.g. scale2), because once a scale
+            # is skipped, all the lower scales are also skipped
+            continue
+
         if raster_translation is None:
             raster_translation = raster_translation_single_scale
         # we set a dummy empty dict for the transformation that will be replaced with the correct transformation for

tests/core/operations/test_transform.py

@@ -11,7 +11,7 @@
 from spatialdata._core.data_extent import are_extents_equal, get_extent
 from spatialdata._core.spatialdata import SpatialData
 from spatialdata._utils import unpad_raster
-from spatialdata.models import PointsModel, ShapesModel, get_axes_names
+from spatialdata.models import Image2DModel, PointsModel, ShapesModel, get_axes_names
 from spatialdata.transformations.operations import (
     align_elements_using_landmarks,
     get_transformation,
@@ -229,6 +229,37 @@ def test_transform_shapes(shapes: SpatialData):
         assert geom_almost_equals(p0["geometry"], p1["geometry"])
 
 
+def test_transform_datatree_scale_handling():
+    """
+    Test the cases in which the lowest and highest scale of the result of a
+    transformed multi-scale image would be zero shape.
+    """
+
+    test_image = Image2DModel.parse(
+        np.ones((1, 10, 10)),
+        dims=("c", "y", "x"),
+        scale_factors=[2, 4],
+        transformations={
+            "cs1": Scale([0.5] * 2, axes=["y", "x"]),
+            "cs2": Scale([0.01] * 2, axes=["y", "x"]),
+        },
+    )
+
+    # check that the transform doesn't raise an error and that it
+    # discards the lowest resolution level
+    test_image_t = transform(test_image, to_coordinate_system="cs1")
+    assert list(test_image.keys()) == ["scale0", "scale1", "scale2"]
+    assert list(test_image_t.keys()) == ["scale0", "scale1"]
+
+    # check that a ValueError is raised when no resolution level
+    # is left after the transformation
+    with pytest.raises(
+        ValueError,
+        match="The transformation leads to zero shaped data even at the highest resolution level",
+    ):
+        transform(test_image, to_coordinate_system="cs2")
+
+
 def test_map_coordinate_systems_single_path(full_sdata: SpatialData):
     scale = Scale([2], axes=("x",))
     translation = Translation([100], axes=("x",))