perspective verified

Author: justincdavis

test/test_transforms_v2.py

@@ -5204,11 +5204,8 @@ def test_image_functional_correctness(self, coefficients, interpolation, fill, m
             image, startpoints=None, endpoints=None, coefficients=coefficients, interpolation=interpolation, fill=fill
         )
         if make_input is make_image_cvcuda:
-            actual = F.cvcuda_to_tensor(actual).to(device="cpu")
-            actual = actual.squeeze(0)
-            # drop the batch dimension
-            image = F.cvcuda_to_tensor(image).to(device="cpu")
-            image = image.squeeze(0)
+            actual = cvcuda_to_pil_compatible_tensor(actual)
+            image = cvcuda_to_pil_compatible_tensor(image)
 
         expected = F.to_image(
             F.perspective(
@@ -5234,7 +5231,7 @@ def test_image_functional_correctness(self, coefficients, interpolation, fill, m
             # visually the results are the same on real images,
             # realistically, the diff is not visible to the human eye
             tolerance = 255 if interpolation is transforms.InterpolationMode.NEAREST else 125
-            torch.testing.assert_close(actual, expected, rtol=0, atol=tolerance)
+            assert_close(actual, expected, rtol=0, atol=tolerance)
 
     def _reference_perspective_bounding_boxes(self, bounding_boxes, *, startpoints, endpoints):
         format = bounding_boxes.format

torchvision/transforms/v2/_geometry.py

@@ -11,7 +11,7 @@
 from torchvision.ops.boxes import box_iou
 from torchvision.transforms.functional import _get_perspective_coeffs
 from torchvision.transforms.v2 import functional as F, InterpolationMode, Transform
-from torchvision.transforms.v2.functional._utils import _FillType
+from torchvision.transforms.v2.functional._utils import _FillType, is_cvcuda_tensor
 
 from ._transform import _RandomApplyTransform
 from ._utils import (
@@ -936,6 +936,8 @@ class RandomPerspective(_RandomApplyTransform):
 
     _v1_transform_cls = _transforms.RandomPerspective
 
+    _transformed_types = _RandomApplyTransform._transformed_types + (is_cvcuda_tensor,)
+
     def __init__(
         self,
         distortion_scale: float = 0.5,