From abe4eec21650418a7daeeea3b90cc4773c3cbb37 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 02:48:47 +0000
Subject: [PATCH 01/10] Implement numpy version of SplitOnGrid

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 .../pathology/transforms/spatial/array.py     | 67 ++++++++++++++-----
 .../transforms/spatial/dictionary.py          |  6 +-
 2 files changed, 55 insertions(+), 18 deletions(-)

diff --git a/monai/apps/pathology/transforms/spatial/array.py b/monai/apps/pathology/transforms/spatial/array.py
index 56927a8033..ef5fce875b 100644
--- a/monai/apps/pathology/transforms/spatial/array.py
+++ b/monai/apps/pathology/transforms/spatial/array.py
@@ -16,6 +16,7 @@
 from numpy.lib.stride_tricks import as_strided
 
 from monai.transforms.transform import Randomizable, Transform
+from monai.utils.enums import TransformBackends
 
 __all__ = ["SplitOnGrid", "TileOnGrid"]
 
@@ -35,6 +36,8 @@ class SplitOnGrid(Transform):
     Note: the shape of the input image is inferred based on the first image used.
     """
 
+    backend = [TransformBackends.TORCH, TransformBackends.NUMPY]
+
     def __init__(
         self, grid_size: Union[int, Tuple[int, int]] = (2, 2), patch_size: Optional[Union[int, Tuple[int, int]]] = None
     ):
@@ -50,17 +53,47 @@ def __init__(
         else:
             self.patch_size = patch_size
 
-    def __call__(self, image: torch.Tensor) -> torch.Tensor:
+    def __call__(self, image: Union[np.ndarray, torch.Tensor]) -> Union[np.ndarray, torch.Tensor]:
         if self.grid_size == (1, 1) and self.patch_size is None:
-            return torch.stack([image])
+            if isinstance(image, torch.Tensor):
+                return torch.stack([image])
+            elif isinstance(image, np.ndarray):
+                return np.stack([image])
+            else:
+                raise ValueError(f"Input type [{type(image)}] is not supported.")
+
         patch_size, steps = self.get_params(image.shape[1:])
-        patches = (
-            image.unfold(1, patch_size[0], steps[0])
-            .unfold(2, patch_size[1], steps[1])
-            .flatten(1, 2)
-            .transpose(0, 1)
-            .contiguous()
-        )
+        patches: Union[np.ndarray, torch.Tensor]
+        if isinstance(image, torch.Tensor):
+            patches = (
+                image.unfold(1, patch_size[0], steps[0])
+                .unfold(2, patch_size[1], steps[1])
+                .flatten(1, 2)
+                .transpose(0, 1)
+                .contiguous()
+            )
+        elif isinstance(image, np.ndarray):
+            x_step, y_step = steps
+            c_stride, x_stride, y_stride = image.strides
+            patches = as_strided(
+                image,
+                shape=(*self.grid_size, 3, patch_size[0], patch_size[1]),
+                strides=(
+                    x_stride * x_step,
+                    y_stride * y_step,
+                    c_stride,
+                    x_stride,
+                    y_stride,
+                ),
+                writeable=False,
+            )
+            # flatten the first two dimensions
+            patches = patches.reshape(np.prod(patches.shape[:2]), *patches.shape[2:])
+            # make it a contiguous array
+            patches = np.ascontiguousarray(patches)
+        else:
+            raise ValueError(f"Input type [{type(image)}] is not supported.")
+
         return patches
 
     def get_params(self, image_size):
@@ -101,6 +134,8 @@ class TileOnGrid(Randomizable, Transform):
 
     """
 
+    backend = [TransformBackends.NUMPY]
+
     def __init__(
         self,
         tile_count: Optional[int] = None,
@@ -177,17 +212,17 @@ def __call__(self, image: np.ndarray) -> np.ndarray:
             )
 
         # extact tiles
-        xstep, ystep = self.step, self.step
-        xsize, ysize = self.tile_size, self.tile_size
-        clen, xlen, ylen = image.shape
-        cstride, xstride, ystride = image.strides
+        x_step, y_step = self.step, self.step
+        x_size, y_size = self.tile_size, self.tile_size
+        c_len, x_len, y_len = image.shape
+        c_stride, x_stride, y_stride = image.strides
         llw = as_strided(
             image,
-            shape=((xlen - xsize) // xstep + 1, (ylen - ysize) // ystep + 1, clen, xsize, ysize),
-            strides=(xstride * xstep, ystride * ystep, cstride, xstride, ystride),
+            shape=((x_len - x_size) // x_step + 1, (y_len - y_size) // y_step + 1, c_len, x_size, y_size),
+            strides=(x_stride * x_step, y_stride * y_step, c_stride, x_stride, y_stride),
             writeable=False,
         )
-        image = llw.reshape(-1, clen, xsize, ysize)
+        image = llw.reshape(-1, c_len, x_size, y_size)
 
         # if keeping all patches
         if self.tile_count is None:
diff --git a/monai/apps/pathology/transforms/spatial/dictionary.py b/monai/apps/pathology/transforms/spatial/dictionary.py
index f998e53c93..8ca00b4987 100644
--- a/monai/apps/pathology/transforms/spatial/dictionary.py
+++ b/monai/apps/pathology/transforms/spatial/dictionary.py
@@ -35,7 +35,7 @@ class SplitOnGridd(MapTransform):
             If it's an integer, the value will be repeated for each dimension.
             The default is (0, 0), where the patch size will be inferred from the grid shape.
 
-    Note: the shape of the input image is infered based on the first image used.
+    Note: the shape of the input image is inferred based on the first image used.
     """
 
     def __init__(
@@ -48,7 +48,9 @@ def __init__(
         super().__init__(keys, allow_missing_keys)
         self.splitter = SplitOnGrid(grid_size=grid_size, patch_size=patch_size)
 
-    def __call__(self, data: Mapping[Hashable, torch.Tensor]) -> Dict[Hashable, torch.Tensor]:
+    def __call__(
+        self, data: Mapping[Hashable, Union[np.ndarray, torch.Tensor]]
+    ) -> Dict[Hashable, Union[np.ndarray, torch.Tensor]]:
         d = dict(data)
         for key in self.key_iterator(d):
             d[key] = self.splitter(d[key])

From 1a5f590e11f3d5ea6d6ed1c1e2134af51627dd09 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 03:02:34 +0000
Subject: [PATCH 02/10] Add test cases for numpy in array tranform

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 tests/test_split_on_grid.py | 32 ++++++++++++++++++++++++++++++--
 1 file changed, 30 insertions(+), 2 deletions(-)

diff --git a/tests/test_split_on_grid.py b/tests/test_split_on_grid.py
index 4893c4c78a..c92ad5f37a 100644
--- a/tests/test_split_on_grid.py
+++ b/tests/test_split_on_grid.py
@@ -52,16 +52,44 @@
 
 
 class TestSplitOnGrid(unittest.TestCase):
+    @parameterized.expand(
+        [
+            TEST_CASE_0,
+            TEST_CASE_1,
+            TEST_CASE_2,
+            TEST_CASE_3,
+            TEST_CASE_4,
+            TEST_CASE_5,
+            TEST_CASE_6,
+            TEST_CASE_7,
+        ]
+    )
+    def test_split_patch_single_call_numpy(self, input_parameters, img, expected):
+        img = img.numpy()
+        expected = expected.numpy()
+        splitter = SplitOnGrid(**input_parameters)
+        output = splitter(img)
+        np.testing.assert_equal(output, expected)
+
+    @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
+    def test_split_patch_multiple_call_numpy(self, input_parameters, img_list, expected_list):
+        splitter = SplitOnGrid(**input_parameters)
+        for img, expected in zip(img_list, expected_list):
+            img = img.numpy()
+            expected = expected.numpy()
+            output = splitter(img)
+            np.testing.assert_equal(output, expected)
+
     @parameterized.expand(
         [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
     )
-    def test_split_pathce_single_call(self, input_parameters, img, expected):
+    def test_split_patch_single_call_torch(self, input_parameters, img, expected):
         splitter = SplitOnGrid(**input_parameters)
         output = splitter(img)
         np.testing.assert_equal(output.numpy(), expected.numpy())
 
     @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
-    def test_split_pathce_multiple_call(self, input_parameters, img_list, expected_list):
+    def test_split_patch_multiple_call_torch(self, input_parameters, img_list, expected_list):
         splitter = SplitOnGrid(**input_parameters)
         for img, expected in zip(img_list, expected_list):
             output = splitter(img)

From 010dd5abe82111f847c6f2bfee76a0b6f236097b Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 03:07:02 +0000
Subject: [PATCH 03/10] Add test cases for numpy in dictionary transform

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 tests/test_split_on_grid_dict.py | 31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)

diff --git a/tests/test_split_on_grid_dict.py b/tests/test_split_on_grid_dict.py
index f22e58515f..606439c1ad 100644
--- a/tests/test_split_on_grid_dict.py
+++ b/tests/test_split_on_grid_dict.py
@@ -49,7 +49,11 @@
 ]
 
 
-TEST_CASE_MC_1 = [{"keys": "image", "grid_size": (2, 1)}, [{"image": A}] * 5, [torch.stack([A1, A2])] * 5]
+TEST_CASE_MC_1 = [
+    {"keys": "image", "grid_size": (2, 1)},
+    [{"image": A}, {"image": A}, {"image": A}],
+    [torch.stack([A1, A2])] * 3,
+]
 
 
 TEST_CASE_MC_2 = [
@@ -63,13 +67,34 @@ class TestSplitOnGridDict(unittest.TestCase):
     @parameterized.expand(
         [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
     )
-    def test_split_pathce_single_call(self, input_parameters, img_dict, expected):
+    def test_split_patch_single_call_numpy(self, input_parameters, img_dict, expected):
+        img_dict_np = {}
+        for k, v in img_dict.items():
+            img_dict_np[k] = v.numpy()
+        splitter = SplitOnGridDict(**input_parameters)
+        output = splitter(img_dict_np)[input_parameters["keys"]]
+        np.testing.assert_equal(output, expected.numpy())
+
+    @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
+    def test_split_patch_multiple_call_numpy(self, input_parameters, img_list, expected_list):
+        splitter = SplitOnGridDict(**input_parameters)
+        for img_dict, expected in zip(img_list, expected_list):
+            img_dict_np = {}
+            for k, v in img_dict.items():
+                img_dict_np[k] = v.numpy()
+            output = splitter(img_dict_np)[input_parameters["keys"]]
+            np.testing.assert_equal(output, expected.numpy())
+
+    @parameterized.expand(
+        [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
+    )
+    def test_split_patch_single_call_torch(self, input_parameters, img_dict, expected):
         splitter = SplitOnGridDict(**input_parameters)
         output = splitter(img_dict)[input_parameters["keys"]]
         np.testing.assert_equal(output.numpy(), expected.numpy())
 
     @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
-    def test_split_pathce_multiple_call(self, input_parameters, img_list, expected_list):
+    def test_split_patch_multiple_call_torch(self, input_parameters, img_list, expected_list):
         splitter = SplitOnGridDict(**input_parameters)
         for img_dict, expected in zip(img_list, expected_list):
             output = splitter(img_dict)[input_parameters["keys"]]

From 19ff1ac74ef54357ff9e9ebe0302421e93832a3c Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 03:09:47 +0000
Subject: [PATCH 04/10] Remove trailing comma

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 tests/test_split_on_grid.py | 11 +----------
 1 file changed, 1 insertion(+), 10 deletions(-)

diff --git a/tests/test_split_on_grid.py b/tests/test_split_on_grid.py
index c92ad5f37a..aa60c5b4bd 100644
--- a/tests/test_split_on_grid.py
+++ b/tests/test_split_on_grid.py
@@ -53,16 +53,7 @@
 
 class TestSplitOnGrid(unittest.TestCase):
     @parameterized.expand(
-        [
-            TEST_CASE_0,
-            TEST_CASE_1,
-            TEST_CASE_2,
-            TEST_CASE_3,
-            TEST_CASE_4,
-            TEST_CASE_5,
-            TEST_CASE_6,
-            TEST_CASE_7,
-        ]
+        [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
     )
     def test_split_patch_single_call_numpy(self, input_parameters, img, expected):
         img = img.numpy()

From b360feea393391913d0fc99f3da1596a682a64d1 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 03:12:21 +0000
Subject: [PATCH 05/10] Add backend to SplitOnGridd and TileOnGridd

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 monai/apps/pathology/transforms/spatial/dictionary.py | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/monai/apps/pathology/transforms/spatial/dictionary.py b/monai/apps/pathology/transforms/spatial/dictionary.py
index 8ca00b4987..81f6b65a22 100644
--- a/monai/apps/pathology/transforms/spatial/dictionary.py
+++ b/monai/apps/pathology/transforms/spatial/dictionary.py
@@ -38,6 +38,8 @@ class SplitOnGridd(MapTransform):
     Note: the shape of the input image is inferred based on the first image used.
     """
 
+    backend = SplitOnGrid.backend
+
     def __init__(
         self,
         keys: KeysCollection,
@@ -81,6 +83,8 @@ class TileOnGridd(Randomizable, MapTransform):
 
     """
 
+    backend = TileOnGrid.backend
+
     def __init__(
         self,
         keys: KeysCollection,

From 7d12ea8310f4b173b09f6eaa6bbb350863119165 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 03:26:34 +0000
Subject: [PATCH 06/10] Remove a trailing comma and rename few variables

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 monai/apps/pathology/transforms/spatial/array.py | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)

diff --git a/monai/apps/pathology/transforms/spatial/array.py b/monai/apps/pathology/transforms/spatial/array.py
index ef5fce875b..27725f2a34 100644
--- a/monai/apps/pathology/transforms/spatial/array.py
+++ b/monai/apps/pathology/transforms/spatial/array.py
@@ -73,18 +73,12 @@ def __call__(self, image: Union[np.ndarray, torch.Tensor]) -> Union[np.ndarray,
                 .contiguous()
             )
         elif isinstance(image, np.ndarray):
-            x_step, y_step = steps
-            c_stride, x_stride, y_stride = image.strides
+            h_step, w_step = steps
+            c_stride, h_stride, w_stride = image.strides
             patches = as_strided(
                 image,
                 shape=(*self.grid_size, 3, patch_size[0], patch_size[1]),
-                strides=(
-                    x_stride * x_step,
-                    y_stride * y_step,
-                    c_stride,
-                    x_stride,
-                    y_stride,
-                ),
+                strides=(h_stride * h_step, w_stride * w_step, c_stride, h_stride, w_stride),
                 writeable=False,
             )
             # flatten the first two dimensions

From b8b44ef0eb6864f75ea9147d8f338424dd2cfef2 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 03:41:57 +0000
Subject: [PATCH 07/10] Add NdarrayOrTensor and remove TileOnGrid backend

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 monai/apps/pathology/transforms/spatial/array.py      | 7 +++----
 monai/apps/pathology/transforms/spatial/dictionary.py | 5 ++---
 2 files changed, 5 insertions(+), 7 deletions(-)

diff --git a/monai/apps/pathology/transforms/spatial/array.py b/monai/apps/pathology/transforms/spatial/array.py
index 27725f2a34..91e8aa94e5 100644
--- a/monai/apps/pathology/transforms/spatial/array.py
+++ b/monai/apps/pathology/transforms/spatial/array.py
@@ -15,6 +15,7 @@
 import torch
 from numpy.lib.stride_tricks import as_strided
 
+from monai.config.type_definitions import NdarrayOrTensor
 from monai.transforms.transform import Randomizable, Transform
 from monai.utils.enums import TransformBackends
 
@@ -53,7 +54,7 @@ def __init__(
         else:
             self.patch_size = patch_size
 
-    def __call__(self, image: Union[np.ndarray, torch.Tensor]) -> Union[np.ndarray, torch.Tensor]:
+    def __call__(self, image: NdarrayOrTensor) -> NdarrayOrTensor:
         if self.grid_size == (1, 1) and self.patch_size is None:
             if isinstance(image, torch.Tensor):
                 return torch.stack([image])
@@ -63,7 +64,7 @@ def __call__(self, image: Union[np.ndarray, torch.Tensor]) -> Union[np.ndarray,
                 raise ValueError(f"Input type [{type(image)}] is not supported.")
 
         patch_size, steps = self.get_params(image.shape[1:])
-        patches: Union[np.ndarray, torch.Tensor]
+        patches: NdarrayOrTensor
         if isinstance(image, torch.Tensor):
             patches = (
                 image.unfold(1, patch_size[0], steps[0])
@@ -128,8 +129,6 @@ class TileOnGrid(Randomizable, Transform):
 
     """
 
-    backend = [TransformBackends.NUMPY]
-
     def __init__(
         self,
         tile_count: Optional[int] = None,
diff --git a/monai/apps/pathology/transforms/spatial/dictionary.py b/monai/apps/pathology/transforms/spatial/dictionary.py
index 81f6b65a22..2959a9b99e 100644
--- a/monai/apps/pathology/transforms/spatial/dictionary.py
+++ b/monai/apps/pathology/transforms/spatial/dictionary.py
@@ -16,6 +16,7 @@
 import torch
 
 from monai.config import KeysCollection
+from monai.config.type_definitions import NdarrayOrTensor
 from monai.transforms.transform import MapTransform, Randomizable
 
 from .array import SplitOnGrid, TileOnGrid
@@ -50,9 +51,7 @@ def __init__(
         super().__init__(keys, allow_missing_keys)
         self.splitter = SplitOnGrid(grid_size=grid_size, patch_size=patch_size)
 
-    def __call__(
-        self, data: Mapping[Hashable, Union[np.ndarray, torch.Tensor]]
-    ) -> Dict[Hashable, Union[np.ndarray, torch.Tensor]]:
+    def __call__(self, data: Mapping[Hashable, NdarrayOrTensor]) -> Dict[Hashable, NdarrayOrTensor]:
         d = dict(data)
         for key in self.key_iterator(d):
             d[key] = self.splitter(d[key])

From 0a57990cb7611f943c5f03cafbf993db4b18aef7 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 04:02:50 +0000
Subject: [PATCH 08/10] Update unittests to use TEST_NDARRAYS

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 tests/test_split_on_grid.py      | 71 +++++++++++++-------------------
 tests/test_split_on_grid_dict.py | 71 ++++++++++++++------------------
 2 files changed, 59 insertions(+), 83 deletions(-)

diff --git a/tests/test_split_on_grid.py b/tests/test_split_on_grid.py
index aa60c5b4bd..a3bf2674f4 100644
--- a/tests/test_split_on_grid.py
+++ b/tests/test_split_on_grid.py
@@ -11,11 +11,11 @@
 
 import unittest
 
-import numpy as np
 import torch
 from parameterized import parameterized
 
 from monai.apps.pathology.transforms import SplitOnGrid
+from tests.utils import TEST_NDARRAYS, assert_allclose
 
 A11 = torch.randn(3, 2, 2)
 A12 = torch.randn(3, 2, 2)
@@ -27,64 +27,51 @@
 A = torch.cat([A1, A2], 1)
 
 TEST_CASE_0 = [{"grid_size": (2, 2)}, A, torch.stack([A11, A12, A21, A22])]
-
 TEST_CASE_1 = [{"grid_size": (2, 1)}, A, torch.stack([A1, A2])]
-
 TEST_CASE_2 = [{"grid_size": (1, 2)}, A1, torch.stack([A11, A12])]
-
 TEST_CASE_3 = [{"grid_size": (1, 2)}, A2, torch.stack([A21, A22])]
-
 TEST_CASE_4 = [{"grid_size": (1, 1), "patch_size": (2, 2)}, A, torch.stack([A11])]
-
 TEST_CASE_5 = [{"grid_size": 1, "patch_size": 4}, A, torch.stack([A])]
-
 TEST_CASE_6 = [{"grid_size": 2, "patch_size": 2}, A, torch.stack([A11, A12, A21, A22])]
-
 TEST_CASE_7 = [{"grid_size": 1}, A, torch.stack([A])]
 
-TEST_CASE_MC_0 = [{"grid_size": (2, 2)}, [A, A], [torch.stack([A11, A12, A21, A22]), torch.stack([A11, A12, A21, A22])]]
-
+TEST_SINGLE = []
+for p in TEST_NDARRAYS:
+    TEST_SINGLE.append([p, *TEST_CASE_0])
+    TEST_SINGLE.append([p, *TEST_CASE_1])
+    TEST_SINGLE.append([p, *TEST_CASE_2])
+    TEST_SINGLE.append([p, *TEST_CASE_3])
+    TEST_SINGLE.append([p, *TEST_CASE_4])
+    TEST_SINGLE.append([p, *TEST_CASE_5])
+    TEST_SINGLE.append([p, *TEST_CASE_6])
+    TEST_SINGLE.append([p, *TEST_CASE_7])
 
+TEST_CASE_MC_0 = [{"grid_size": (2, 2)}, [A, A], [torch.stack([A11, A12, A21, A22]), torch.stack([A11, A12, A21, A22])]]
 TEST_CASE_MC_1 = [{"grid_size": (2, 1)}, [A] * 5, [torch.stack([A1, A2])] * 5]
-
-
 TEST_CASE_MC_2 = [{"grid_size": (1, 2)}, [A1, A2], [torch.stack([A11, A12]), torch.stack([A21, A22])]]
 
+TEST_MULTIPLE = []
+for p in TEST_NDARRAYS:
+    TEST_MULTIPLE.append([p, *TEST_CASE_MC_0])
+    TEST_MULTIPLE.append([p, *TEST_CASE_MC_1])
+    TEST_MULTIPLE.append([p, *TEST_CASE_MC_2])
 
-class TestSplitOnGrid(unittest.TestCase):
-    @parameterized.expand(
-        [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
-    )
-    def test_split_patch_single_call_numpy(self, input_parameters, img, expected):
-        img = img.numpy()
-        expected = expected.numpy()
-        splitter = SplitOnGrid(**input_parameters)
-        output = splitter(img)
-        np.testing.assert_equal(output, expected)
-
-    @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
-    def test_split_patch_multiple_call_numpy(self, input_parameters, img_list, expected_list):
-        splitter = SplitOnGrid(**input_parameters)
-        for img, expected in zip(img_list, expected_list):
-            img = img.numpy()
-            expected = expected.numpy()
-            output = splitter(img)
-            np.testing.assert_equal(output, expected)
 
-    @parameterized.expand(
-        [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
-    )
-    def test_split_patch_single_call_torch(self, input_parameters, img, expected):
+class TestSplitOnGrid(unittest.TestCase):
+    @parameterized.expand(TEST_SINGLE)
+    def test_split_patch_single_call(self, in_type, input_parameters, image, expected):
+        input_image = in_type(image)
         splitter = SplitOnGrid(**input_parameters)
-        output = splitter(img)
-        np.testing.assert_equal(output.numpy(), expected.numpy())
+        output = splitter(input_image)
+        assert_allclose(output, expected, type_test=False)
 
-    @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
-    def test_split_patch_multiple_call_torch(self, input_parameters, img_list, expected_list):
+    @parameterized.expand(TEST_MULTIPLE)
+    def test_split_patch_multiple_call(self, in_type, input_parameters, img_list, expected_list):
         splitter = SplitOnGrid(**input_parameters)
-        for img, expected in zip(img_list, expected_list):
-            output = splitter(img)
-            np.testing.assert_equal(output.numpy(), expected.numpy())
+        for image, expected in zip(img_list, expected_list):
+            input_image = in_type(image)
+            output = splitter(input_image)
+            assert_allclose(output, expected, type_test=False)
 
 
 if __name__ == "__main__":
diff --git a/tests/test_split_on_grid_dict.py b/tests/test_split_on_grid_dict.py
index 606439c1ad..5f3e442640 100644
--- a/tests/test_split_on_grid_dict.py
+++ b/tests/test_split_on_grid_dict.py
@@ -11,11 +11,11 @@
 
 import unittest
 
-import numpy as np
 import torch
 from parameterized import parameterized
 
 from monai.apps.pathology.transforms import SplitOnGridDict
+from tests.utils import TEST_NDARRAYS, assert_allclose
 
 A11 = torch.randn(3, 2, 2)
 A12 = torch.randn(3, 2, 2)
@@ -27,78 +27,67 @@
 A = torch.cat([A1, A2], 1)
 
 TEST_CASE_0 = [{"keys": "image", "grid_size": (2, 2)}, {"image": A}, torch.stack([A11, A12, A21, A22])]
-
 TEST_CASE_1 = [{"keys": "image", "grid_size": (2, 1)}, {"image": A}, torch.stack([A1, A2])]
-
 TEST_CASE_2 = [{"keys": "image", "grid_size": (1, 2)}, {"image": A1}, torch.stack([A11, A12])]
-
 TEST_CASE_3 = [{"keys": "image", "grid_size": (1, 2)}, {"image": A2}, torch.stack([A21, A22])]
-
 TEST_CASE_4 = [{"keys": "image", "grid_size": (1, 1), "patch_size": (2, 2)}, {"image": A}, torch.stack([A11])]
-
 TEST_CASE_5 = [{"keys": "image", "grid_size": 1, "patch_size": 4}, {"image": A}, torch.stack([A])]
-
 TEST_CASE_6 = [{"keys": "image", "grid_size": 2, "patch_size": 2}, {"image": A}, torch.stack([A11, A12, A21, A22])]
-
 TEST_CASE_7 = [{"keys": "image", "grid_size": 1}, {"image": A}, torch.stack([A])]
 
+TEST_SINGLE = []
+for p in TEST_NDARRAYS:
+    TEST_SINGLE.append([p, *TEST_CASE_0])
+    TEST_SINGLE.append([p, *TEST_CASE_1])
+    TEST_SINGLE.append([p, *TEST_CASE_2])
+    TEST_SINGLE.append([p, *TEST_CASE_3])
+    TEST_SINGLE.append([p, *TEST_CASE_4])
+    TEST_SINGLE.append([p, *TEST_CASE_5])
+    TEST_SINGLE.append([p, *TEST_CASE_6])
+    TEST_SINGLE.append([p, *TEST_CASE_7])
+
 TEST_CASE_MC_0 = [
     {"keys": "image", "grid_size": (2, 2)},
     [{"image": A}, {"image": A}],
     [torch.stack([A11, A12, A21, A22]), torch.stack([A11, A12, A21, A22])],
 ]
-
-
 TEST_CASE_MC_1 = [
     {"keys": "image", "grid_size": (2, 1)},
     [{"image": A}, {"image": A}, {"image": A}],
     [torch.stack([A1, A2])] * 3,
 ]
-
-
 TEST_CASE_MC_2 = [
     {"keys": "image", "grid_size": (1, 2)},
     [{"image": A1}, {"image": A2}],
     [torch.stack([A11, A12]), torch.stack([A21, A22])],
 ]
 
+TEST_MULTIPLE = []
+for p in TEST_NDARRAYS:
+    TEST_MULTIPLE.append([p, *TEST_CASE_MC_0])
+    TEST_MULTIPLE.append([p, *TEST_CASE_MC_1])
+    TEST_MULTIPLE.append([p, *TEST_CASE_MC_2])
+
 
 class TestSplitOnGridDict(unittest.TestCase):
-    @parameterized.expand(
-        [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
-    )
-    def test_split_patch_single_call_numpy(self, input_parameters, img_dict, expected):
-        img_dict_np = {}
+    @parameterized.expand(TEST_SINGLE)
+    def test_split_patch_single_call(self, in_type, input_parameters, img_dict, expected):
+        input_dict = {}
         for k, v in img_dict.items():
-            img_dict_np[k] = v.numpy()
+            input_dict[k] = in_type(v)
         splitter = SplitOnGridDict(**input_parameters)
-        output = splitter(img_dict_np)[input_parameters["keys"]]
-        np.testing.assert_equal(output, expected.numpy())
+        output = splitter(input_dict)[input_parameters["keys"]]
+        assert_allclose(output, expected, type_test=False)
 
-    @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
-    def test_split_patch_multiple_call_numpy(self, input_parameters, img_list, expected_list):
+    @parameterized.expand(TEST_MULTIPLE)
+    def test_split_patch_multiple_call(self, in_type, input_parameters, img_list, expected_list):
         splitter = SplitOnGridDict(**input_parameters)
         for img_dict, expected in zip(img_list, expected_list):
-            img_dict_np = {}
+            input_dict = {}
             for k, v in img_dict.items():
-                img_dict_np[k] = v.numpy()
-            output = splitter(img_dict_np)[input_parameters["keys"]]
-            np.testing.assert_equal(output, expected.numpy())
-
-    @parameterized.expand(
-        [TEST_CASE_0, TEST_CASE_1, TEST_CASE_2, TEST_CASE_3, TEST_CASE_4, TEST_CASE_5, TEST_CASE_6, TEST_CASE_7]
-    )
-    def test_split_patch_single_call_torch(self, input_parameters, img_dict, expected):
-        splitter = SplitOnGridDict(**input_parameters)
-        output = splitter(img_dict)[input_parameters["keys"]]
-        np.testing.assert_equal(output.numpy(), expected.numpy())
-
-    @parameterized.expand([TEST_CASE_MC_0, TEST_CASE_MC_1, TEST_CASE_MC_2])
-    def test_split_patch_multiple_call_torch(self, input_parameters, img_list, expected_list):
-        splitter = SplitOnGridDict(**input_parameters)
-        for img_dict, expected in zip(img_list, expected_list):
-            output = splitter(img_dict)[input_parameters["keys"]]
-            np.testing.assert_equal(output.numpy(), expected.numpy())
+                input_dict[k] = in_type(v)
+            output = splitter(input_dict)[input_parameters["keys"]]
+            assert_allclose(output, expected, type_test=False)
 
 
 if __name__ == "__main__":

From 6db973661e6f99b3b5e2fe2b9ff11b20f1053f90 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 04:08:12 +0000
Subject: [PATCH 09/10] Remove torch import

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 monai/apps/pathology/transforms/spatial/dictionary.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/monai/apps/pathology/transforms/spatial/dictionary.py b/monai/apps/pathology/transforms/spatial/dictionary.py
index 2959a9b99e..e1a483c96f 100644
--- a/monai/apps/pathology/transforms/spatial/dictionary.py
+++ b/monai/apps/pathology/transforms/spatial/dictionary.py
@@ -13,7 +13,6 @@
 from typing import Any, Dict, Hashable, List, Mapping, Optional, Tuple, Union
 
 import numpy as np
-import torch
 
 from monai.config import KeysCollection
 from monai.config.type_definitions import NdarrayOrTensor

From 8defcd2c8fe46496abc97e113554779ec4c091d8 Mon Sep 17 00:00:00 2001
From: Behrooz <3968947+drbeh@users.noreply.github.com>
Date: Mon, 22 Nov 2021 15:43:31 +0000
Subject: [PATCH 10/10] Remove TileOnGridd backend

Signed-off-by: Behrooz <3968947+drbeh@users.noreply.github.com>
---
 monai/apps/pathology/transforms/spatial/dictionary.py | 2 --
 1 file changed, 2 deletions(-)

diff --git a/monai/apps/pathology/transforms/spatial/dictionary.py b/monai/apps/pathology/transforms/spatial/dictionary.py
index e1a483c96f..32df2cae3c 100644
--- a/monai/apps/pathology/transforms/spatial/dictionary.py
+++ b/monai/apps/pathology/transforms/spatial/dictionary.py
@@ -81,8 +81,6 @@ class TileOnGridd(Randomizable, MapTransform):
 
     """
 
-    backend = TileOnGrid.backend
-
     def __init__(
         self,
         keys: KeysCollection,