Infer object-dtype for if extra_property depends on mask size in regionprops_table

[elena-pascal]

Description

Attempt to fix #7129 By using regions of different sizes to do the dtype check.

Checklist

• A descriptive but concise pull request title
• Unit tests I would love to do but for the life of me cannot get scikit-image to build correctly on this system
• Contribution guide is followed

Release note

Summarize the introduced changes in the code block below in one or a few sentences. The
summary will be included in the next release notes automatically:

Allow `extra_properties` of non-equal lengths to be passed correctly to 
`skimage.measure.regionprops_table`.

[stefanv]

@lagru @jni If you have a moment to review, it would be much appreciated.

[elena-pascal]

Sorry folks, I think it's a bit weedy and clearly my solution is failing the tests.

[lagru]

Thanks for taking a shot at this. I had to debug this locally. And to be honest, while doing so already came up with a possible solution. How about this patch?

Patch

Subject: [PATCH] Allow different extra prop shapes in regionprops_table

When the extra_property returns different shapes depending on the mask
shape, the infered dtype should be `np.object_`. The new approach
creates two masks with different shapes in a hopefully more explicit 
way.
---
Index: skimage/measure/tests/test_regionprops.py
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/skimage/measure/tests/test_regionprops.py b/skimage/measure/tests/test_regionprops.py
--- a/skimage/measure/tests/test_regionprops.py	(revision c2d31fbb2706763afba988c4c49bfbf93d24a5e2)
+++ b/skimage/measure/tests/test_regionprops.py	(date 1694860912577)
@@ -1280,6 +1280,11 @@
     return np.median(image_intensity[regionmask])
 
 
+def bbox_list(regionmask):
+    """Extra property who's output shape is dependent on mask shape."""
+    return [1] * regionmask.shape[1]
+
+
 def too_many_args(regionmask, image_intensity, superfluous):
     return 1
 
@@ -1337,14 +1342,19 @@
 
 
 def test_extra_properties_table():
-    out = regionprops_table(SAMPLE_MULTIPLE,
-                            intensity_image=INTENSITY_SAMPLE_MULTIPLE,
-                            properties=('label',),
-                            extra_properties=(intensity_median, pixelcount)
-                            )
+    out = regionprops_table(
+        SAMPLE_MULTIPLE,
+        intensity_image=INTENSITY_SAMPLE_MULTIPLE,
+        properties=('label',),
+        extra_properties=(intensity_median, pixelcount, bbox_list)
+    )
     assert_array_almost_equal(out['intensity_median'], np.array([2., 4.]))
     assert_array_equal(out['pixelcount'], np.array([10, 2]))
 
+    assert out["bbox_list"].dtype == np.object_
+    assert out["bbox_list"][0] == [1] * 10
+    assert out["bbox_list"][1] == [1] * 1
+
 
 def test_multichannel():
     """Test that computing multichannel properties works."""
Index: skimage/measure/_regionprops.py
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/skimage/measure/_regionprops.py b/skimage/measure/_regionprops.py
--- a/skimage/measure/_regionprops.py	(revision c2d31fbb2706763afba988c4c49bfbf93d24a5e2)
+++ b/skimage/measure/_regionprops.py	(date 1694860630250)
@@ -194,20 +194,17 @@
     dtype : NumPy data type
         The data type of the returned property.
     """
-    labels = [1, 2]
-    sample = np.zeros((3,) * ndim, dtype=np.intp)
-
-    regions = [(slice(0,),) * ndim, (slice(1, None),) * ndim]
-    sample[regions[0]] = labels[0]
-    sample[regions[1]] = labels[1]
-
-    propmasks = [(sample[region] == label) for (label, region) in zip(labels, regions)]
+    mask_1 = np.ones((1,) * ndim, dtype=bool)
+    mask_1 = np.pad(mask_1, (0, 1), constant_values=False)
+    mask_2 = np.ones((2,) * ndim, dtype=bool)
+    mask_2 = np.pad(mask_2, (1, 0), constant_values=False)
+    propmasks = [mask_1, mask_2]
 
     rng = np.random.default_rng()
 
     if intensity and _infer_number_of_required_args(func) == 2:
         def _func(mask):
-            return func(mask, rng.random(sample.shape))
+            return func(mask, rng.random(mask.shape))
     else:
         _func = func
     props1, props2 = map(_func, propmasks)

Please, let me know if I should push the patch directly or if you want to improve or iterate on it. :)

[elena-pascal]

The patch looks good to me. Thank you for untangling this :)

[lagru]

Looks good to me. The CI failures should be unrelated and addressed in #7140.

Thanks @elena-pascal for collaborating! :)

[mkcor]

Thanks both for working on this! I just spotted these details; I'll try the code locally now.

[elena-pascal]

Ah ok, I accidently triggered the CI again. Sorry polar bears

[lagru]

Thanks!