7951 anchors are not centered on grid cells (#8475)

This pull request resolves issue #7951 by correctly centering anchor
boxes within their grid cells.

### Description

**Changes Made:**
- Modified the anchor generation logic in
`monai/apps/detection/utils/anchor_utils.py` to add a `stride // 2`
offset.
- Refactored the corresponding unit test in
`tests/apps/detection/utils/test_anchor_box.py` to validate this new,
correct behavior against the torchvision baseline by accounting for the
offset.

Fixes #7951

### Types of changes
<!--- Put an `x` in all the boxes that apply, and remove the not
applicable items -->
- [ ] Non-breaking change (fix or new feature that would not break
existing functionality).
- [x] Breaking change (fix or new feature that would cause existing
functionality to change).
- [x] New tests added to cover the changes.
- [x] Integration tests passed locally by running `./runtests.sh -f -u
--net --coverage`.
- [x] Quick tests passed locally by running `./runtests.sh --quick
--unittests --disttests`.
- [ ] In-line docstrings updated.
- [ ] Documentation updated, tested `make html` command in the `docs/`
folder.
- [ ]

---------

Signed-off-by: hachoj <hjchojnowski@gmail.com>
Co-authored-by: Eric Kerfoot <17726042+ericspod@users.noreply.github.com>

Author: hachoj

monai/apps/detection/utils/anchor_utils.py

@@ -253,7 +253,7 @@ def grid_anchors(self, grid_sizes: list[list[int]], strides: list[list[Tensor]])
             # compute anchor centers regarding to the image.
             # shifts_centers is [x_center, y_center] or [x_center, y_center, z_center]
             shifts_centers = [
-                torch.arange(0, size[axis], dtype=torch.int32, device=device) * stride[axis]
+                torch.arange(0, size[axis], dtype=torch.int32, device=device) * stride[axis] + stride[axis] // 2
                 for axis in range(self.spatial_dims)
             ]
 

tests/apps/detection/utils/test_anchor_box.py

@@ -44,9 +44,9 @@ class TestAnchorGenerator(unittest.TestCase):
     @parameterized.expand(TEST_CASES_2D)
     def test_anchor_2d(self, input_param, image_shape, feature_maps_shapes):
         torch_anchor_utils, _ = optional_import("torchvision.models.detection.anchor_utils")
-        image_list, _ = optional_import("torchvision.models.detection.image_list")
 
-        # test it behaves the same with torchvision for 2d
+        # test it behaves for new functionality of centered anchors
+        # pytorch does not follow this functionality
         anchor = AnchorGenerator(**input_param, indexing="xy")
         anchor_ref = torch_anchor_utils.AnchorGenerator(**input_param)
         for a, a_f in zip(anchor.cell_anchors, anchor_ref.cell_anchors):
@@ -56,15 +56,18 @@ def test_anchor_2d(self, input_param, image_shape, feature_maps_shapes):
 
         grid_sizes = [[2, 2], [1, 1]]
         strides = [[torch.tensor(1), torch.tensor(2)], [torch.tensor(2), torch.tensor(4)]]
-        for a, a_f in zip(anchor.grid_anchors(grid_sizes, strides), anchor_ref.grid_anchors(grid_sizes, strides)):
-            assert_allclose(a, a_f, type_test=True, device_test=False, atol=1e-3)
 
-        images = torch.rand(image_shape)
-        feature_maps = tuple(torch.rand(fs) for fs in feature_maps_shapes)
-        result = anchor(images, feature_maps)
-        result_ref = anchor_ref(image_list.ImageList(images, ([123, 122],)), feature_maps)
-        for a, a_f in zip(result, result_ref):
-            assert_allclose(a, a_f, type_test=True, device_test=False, atol=0.1)
+        monai_anchors = anchor.grid_anchors(grid_sizes, strides)
+        torchvision_anchors = anchor_ref.grid_anchors(grid_sizes, strides)
+
+        for a, a_f, s in zip(monai_anchors, torchvision_anchors, strides):
+            stride_y, stride_x = s
+
+            offset_x = stride_x // 2
+            offset_y = stride_y // 2
+            offset = torch.tensor([offset_x, offset_y, offset_x, offset_y], dtype=a_f.dtype, device=a_f.device)
+
+            assert_allclose(a, a_f + offset, type_test=True, device_test=False, atol=1e-3)
 
     @parameterized.expand(TEST_CASES_2D)
     def test_script_2d(self, input_param, image_shape, feature_maps_shapes):