clamp bounding boxes in some geometry kernels (#7215)

Co-authored-by: vfdev-5 <[email protected]>

Author: pmeier

test/prototype_transforms_kernel_infos.py

@@ -108,7 +108,7 @@ def float32_vs_uint8_pixel_difference(atol=1, mae=False):
     }
 
 
-def scripted_vs_eager_double_pixel_difference(device, atol=1e-6, rtol=1e-6):
+def scripted_vs_eager_float64_tolerances(device, atol=1e-6, rtol=1e-6):
     return {
         (("TestKernels", "test_scripted_vs_eager"), torch.float64, device): {"atol": atol, "rtol": rtol, "mae": False},
     }
@@ -211,10 +211,12 @@ def reference_horizontal_flip_bounding_box(bounding_box, *, format, spatial_size
             [-1, 0, spatial_size[1]],
             [0, 1, 0],
         ],
-        dtype="float32",
+        dtype="float64" if bounding_box.dtype == torch.float64 else "float32",
     )
 
-    expected_bboxes = reference_affine_bounding_box_helper(bounding_box, format=format, affine_matrix=affine_matrix)
+    expected_bboxes = reference_affine_bounding_box_helper(
+        bounding_box, format=format, spatial_size=spatial_size, affine_matrix=affine_matrix
+    )
 
     return expected_bboxes
 
@@ -322,7 +324,7 @@ def reference_inputs_resize_image_tensor():
 def sample_inputs_resize_bounding_box():
     for bounding_box_loader in make_bounding_box_loaders():
         for size in _get_resize_sizes(bounding_box_loader.spatial_size):
-            yield ArgsKwargs(bounding_box_loader, size=size, spatial_size=bounding_box_loader.spatial_size)
+            yield ArgsKwargs(bounding_box_loader, spatial_size=bounding_box_loader.spatial_size, size=size)
 
 
 def sample_inputs_resize_mask():
@@ -344,19 +346,20 @@ def reference_resize_bounding_box(bounding_box, *, spatial_size, size, max_size=
             [new_width / old_width, 0, 0],
             [0, new_height / old_height, 0],
         ],
-        dtype="float32",
+        dtype="float64" if bounding_box.dtype == torch.float64 else "float32",
     )
 
     expected_bboxes = reference_affine_bounding_box_helper(
-        bounding_box, format=datapoints.BoundingBoxFormat.XYXY, affine_matrix=affine_matrix
+        bounding_box,
+        format=bounding_box.format,
+        spatial_size=(new_height, new_width),
+        affine_matrix=affine_matrix,
     )
     return expected_bboxes, (new_height, new_width)
 
 
 def reference_inputs_resize_bounding_box():
-    for bounding_box_loader in make_bounding_box_loaders(
-        formats=[datapoints.BoundingBoxFormat.XYXY], extra_dims=((), (4,))
-    ):
+    for bounding_box_loader in make_bounding_box_loaders(extra_dims=((), (4,))):
         for size in _get_resize_sizes(bounding_box_loader.spatial_size):
             yield ArgsKwargs(bounding_box_loader, size=size, spatial_size=bounding_box_loader.spatial_size)
 
@@ -543,14 +546,17 @@ def _compute_affine_matrix(angle, translate, scale, shear, center):
     return true_matrix
 
 
-def reference_affine_bounding_box_helper(bounding_box, *, format, affine_matrix):
-    def transform(bbox, affine_matrix_, format_):
+def reference_affine_bounding_box_helper(bounding_box, *, format, spatial_size, affine_matrix):
+    def transform(bbox, affine_matrix_, format_, spatial_size_):
         # Go to float before converting to prevent precision loss in case of CXCYWH -> XYXY and W or H is 1
         in_dtype = bbox.dtype
         if not torch.is_floating_point(bbox):
             bbox = bbox.float()
         bbox_xyxy = F.convert_format_bounding_box(
-            bbox, old_format=format_, new_format=datapoints.BoundingBoxFormat.XYXY, inplace=True
+            bbox.as_subclass(torch.Tensor),
+            old_format=format_,
+            new_format=datapoints.BoundingBoxFormat.XYXY,
+            inplace=True,
         )
         points = np.array(
             [
@@ -573,12 +579,15 @@ def transform(bbox, affine_matrix_, format_):
         out_bbox = F.convert_format_bounding_box(
             out_bbox, old_format=datapoints.BoundingBoxFormat.XYXY, new_format=format_, inplace=True
         )
-        return out_bbox.to(dtype=in_dtype)
+        # It is important to clamp before casting, especially for CXCYWH format, dtype=int64
+        out_bbox = F.clamp_bounding_box(out_bbox, format=format_, spatial_size=spatial_size_)
+        out_bbox = out_bbox.to(dtype=in_dtype)
+        return out_bbox
 
     if bounding_box.ndim < 2:
         bounding_box = [bounding_box]
 
-    expected_bboxes = [transform(bbox, affine_matrix, format) for bbox in bounding_box]
+    expected_bboxes = [transform(bbox, affine_matrix, format, spatial_size) for bbox in bounding_box]
     if len(expected_bboxes) > 1:
         expected_bboxes = torch.stack(expected_bboxes)
     else:
@@ -594,7 +603,9 @@ def reference_affine_bounding_box(bounding_box, *, format, spatial_size, angle,
     affine_matrix = _compute_affine_matrix(angle, translate, scale, shear, center)
     affine_matrix = affine_matrix[:2, :]
 
-    expected_bboxes = reference_affine_bounding_box_helper(bounding_box, format=format, affine_matrix=affine_matrix)
+    expected_bboxes = reference_affine_bounding_box_helper(
+        bounding_box, format=format, spatial_size=spatial_size, affine_matrix=affine_matrix
+    )
 
     return expected_bboxes
 
@@ -643,9 +654,6 @@ def sample_inputs_affine_video():
             sample_inputs_fn=sample_inputs_affine_bounding_box,
             reference_fn=reference_affine_bounding_box,
             reference_inputs_fn=reference_inputs_affine_bounding_box,
-            closeness_kwargs={
-                (("TestKernels", "test_against_reference"), torch.int64, "cpu"): dict(atol=1, rtol=0),
-            },
             test_marks=[
                 xfail_jit_python_scalar_arg("shear"),
             ],
@@ -729,10 +737,12 @@ def reference_vertical_flip_bounding_box(bounding_box, *, format, spatial_size):
             [1, 0, 0],
             [0, -1, spatial_size[0]],
         ],
-        dtype="float32",
+        dtype="float64" if bounding_box.dtype == torch.float64 else "float32",
     )
 
-    expected_bboxes = reference_affine_bounding_box_helper(bounding_box, format=format, affine_matrix=affine_matrix)
+    expected_bboxes = reference_affine_bounding_box_helper(
+        bounding_box, format=format, spatial_size=spatial_size, affine_matrix=affine_matrix
+    )
 
     return expected_bboxes
 
@@ -806,6 +816,43 @@ def sample_inputs_rotate_bounding_box():
         )
 
 
+def reference_inputs_rotate_bounding_box():
+    for bounding_box_loader, angle in itertools.product(
+        make_bounding_box_loaders(extra_dims=((), (4,))), _ROTATE_ANGLES
+    ):
+        yield ArgsKwargs(
+            bounding_box_loader,
+            format=bounding_box_loader.format,
+            spatial_size=bounding_box_loader.spatial_size,
+            angle=angle,
+        )
+
+    # TODO: add samples with expand=True and center
+
+
+def reference_rotate_bounding_box(bounding_box, *, format, spatial_size, angle, expand=False, center=None):
+
+    if center is None:
+        center = [spatial_size[1] * 0.5, spatial_size[0] * 0.5]
+
+    a = np.cos(angle * np.pi / 180.0)
+    b = np.sin(angle * np.pi / 180.0)
+    cx = center[0]
+    cy = center[1]
+    affine_matrix = np.array(
+        [
+            [a, b, cx - cx * a - b * cy],
+            [-b, a, cy + cx * b - a * cy],
+        ],
+        dtype="float64" if bounding_box.dtype == torch.float64 else "float32",
+    )
+
+    expected_bboxes = reference_affine_bounding_box_helper(
+        bounding_box, format=format, spatial_size=spatial_size, affine_matrix=affine_matrix
+    )
+    return expected_bboxes, spatial_size
+
+
 def sample_inputs_rotate_mask():
     for mask_loader in make_mask_loaders(sizes=["random"], num_categories=["random"], num_objects=["random"]):
         yield ArgsKwargs(mask_loader, angle=15.0)
@@ -834,9 +881,11 @@ def sample_inputs_rotate_video():
         KernelInfo(
             F.rotate_bounding_box,
             sample_inputs_fn=sample_inputs_rotate_bounding_box,
+            reference_fn=reference_rotate_bounding_box,
+            reference_inputs_fn=reference_inputs_rotate_bounding_box,
             closeness_kwargs={
-                **scripted_vs_eager_double_pixel_difference("cpu", atol=1e-5, rtol=1e-5),
-                **scripted_vs_eager_double_pixel_difference("cuda", atol=1e-5, rtol=1e-5),
+                **scripted_vs_eager_float64_tolerances("cpu", atol=1e-6, rtol=1e-6),
+                **scripted_vs_eager_float64_tolerances("cuda", atol=1e-5, rtol=1e-5),
             },
         ),
         KernelInfo(
@@ -897,17 +946,19 @@ def sample_inputs_crop_video():
 
 
 def reference_crop_bounding_box(bounding_box, *, format, top, left, height, width):
-
     affine_matrix = np.array(
         [
             [1, 0, -left],
             [0, 1, -top],
         ],
-        dtype="float32",
+        dtype="float64" if bounding_box.dtype == torch.float64 else "float32",
     )
 
-    expected_bboxes = reference_affine_bounding_box_helper(bounding_box, format=format, affine_matrix=affine_matrix)
-    return expected_bboxes, (height, width)
+    spatial_size = (height, width)
+    expected_bboxes = reference_affine_bounding_box_helper(
+        bounding_box, format=format, spatial_size=spatial_size, affine_matrix=affine_matrix
+    )
+    return expected_bboxes, spatial_size
 
 
 def reference_inputs_crop_bounding_box():
@@ -1119,13 +1170,15 @@ def reference_pad_bounding_box(bounding_box, *, format, spatial_size, padding, p
             [1, 0, left],
             [0, 1, top],
         ],
-        dtype="float32",
+        dtype="float64" if bounding_box.dtype == torch.float64 else "float32",
     )
 
     height = spatial_size[0] + top + bottom
     width = spatial_size[1] + left + right
 
-    expected_bboxes = reference_affine_bounding_box_helper(bounding_box, format=format, affine_matrix=affine_matrix)
+    expected_bboxes = reference_affine_bounding_box_helper(
+        bounding_box, format=format, spatial_size=(height, width), affine_matrix=affine_matrix
+    )
     return expected_bboxes, (height, width)
 
 
@@ -1225,14 +1278,16 @@ def sample_inputs_perspective_bounding_box():
         yield ArgsKwargs(
             bounding_box_loader,
             format=bounding_box_loader.format,
+            spatial_size=bounding_box_loader.spatial_size,
             startpoints=None,
             endpoints=None,
             coefficients=_PERSPECTIVE_COEFFS[0],
         )
 
     format = datapoints.BoundingBoxFormat.XYXY
+    loader = make_bounding_box_loader(format=format)
     yield ArgsKwargs(
-        make_bounding_box_loader(format=format), format=format, startpoints=_STARTPOINTS, endpoints=_ENDPOINTS
+        loader, format=format, spatial_size=loader.spatial_size, startpoints=_STARTPOINTS, endpoints=_ENDPOINTS
     )
 
 
@@ -1269,13 +1324,17 @@ def sample_inputs_perspective_video():
                 **pil_reference_pixel_difference(2, mae=True),
                 **cuda_vs_cpu_pixel_difference(),
                 **float32_vs_uint8_pixel_difference(),
-                **scripted_vs_eager_double_pixel_difference("cpu", atol=1e-5, rtol=1e-5),
-                **scripted_vs_eager_double_pixel_difference("cuda", atol=1e-5, rtol=1e-5),
+                **scripted_vs_eager_float64_tolerances("cpu", atol=1e-5, rtol=1e-5),
+                **scripted_vs_eager_float64_tolerances("cuda", atol=1e-5, rtol=1e-5),
             },
         ),
         KernelInfo(
             F.perspective_bounding_box,
             sample_inputs_fn=sample_inputs_perspective_bounding_box,
+            closeness_kwargs={
+                **scripted_vs_eager_float64_tolerances("cpu", atol=1e-6, rtol=1e-6),
+                **scripted_vs_eager_float64_tolerances("cuda", atol=1e-6, rtol=1e-6),
+            },
         ),
         KernelInfo(
             F.perspective_mask,
@@ -1292,8 +1351,8 @@ def sample_inputs_perspective_video():
             sample_inputs_fn=sample_inputs_perspective_video,
             closeness_kwargs={
                 **cuda_vs_cpu_pixel_difference(),
-                **scripted_vs_eager_double_pixel_difference("cpu", atol=1e-5, rtol=1e-5),
-                **scripted_vs_eager_double_pixel_difference("cuda", atol=1e-5, rtol=1e-5),
+                **scripted_vs_eager_float64_tolerances("cpu", atol=1e-5, rtol=1e-5),
+                **scripted_vs_eager_float64_tolerances("cuda", atol=1e-5, rtol=1e-5),
             },
         ),
     ]
@@ -1331,6 +1390,7 @@ def sample_inputs_elastic_bounding_box():
         yield ArgsKwargs(
             bounding_box_loader,
             format=bounding_box_loader.format,
+            spatial_size=bounding_box_loader.spatial_size,
             displacement=displacement,
         )
 

test/test_prototype_transforms.py

@@ -146,7 +146,7 @@ class TestSmoke:
             (transforms.RandomZoomOut(p=1.0), None),
             (transforms.Resize([16, 16], antialias=True), None),
             (transforms.ScaleJitter((16, 16), scale_range=(0.8, 1.2)), None),
-            (transforms.ClampBoundingBoxes(), None),
+            (transforms.ClampBoundingBox(), None),
             (transforms.ConvertBoundingBoxFormat(datapoints.BoundingBoxFormat.CXCYWH), None),
             (transforms.ConvertDtype(), None),
             (transforms.GaussianBlur(kernel_size=3), None),