Unified inputs for F.rotate (#2495)

* Added code for F_t.rotate with test
- updated F.affine tests

* Rotate test tolerance to 2%

* Fixes failing test

* Optimized _expanded_affine_grid with a single matmul op

* Recoded _compute_output_size

Author: vfdev-5

test/test_functional_tensor.py

@@ -435,7 +435,7 @@ def test_affine(self):
                 )
         # 3) Test translation
         test_configs = [
-            [10, 12], (12, 13)
+            [10, 12], (-12, -13)
         ]
         for t in test_configs:
             for fn in [F.affine, scripted_affine]:
@@ -447,21 +447,21 @@ def test_affine(self):
         test_configs = [
             (45, [5, 6], 1.0, [0.0, 0.0]),
             (33, (5, -4), 1.0, [0.0, 0.0]),
-            (45, [5, 4], 1.2, [0.0, 0.0]),
-            (33, (4, 8), 2.0, [0.0, 0.0]),
+            (45, [-5, 4], 1.2, [0.0, 0.0]),
+            (33, (-4, -8), 2.0, [0.0, 0.0]),
             (85, (10, -10), 0.7, [0.0, 0.0]),
             (0, [0, 0], 1.0, [35.0, ]),
             (25, [0, 0], 1.2, [0.0, 15.0]),
-            (45, [10, 0], 0.7, [2.0, 5.0]),
-            (45, [10, -10], 1.2, [4.0, 5.0]),
+            (45, [-10, 0], 0.7, [2.0, 5.0]),
+            (45, [-10, -10], 1.2, [4.0, 5.0]),
         ]
         for r in [0, ]:
             for a, t, s, sh in test_configs:
+                out_pil_img = F.affine(pil_img, angle=a, translate=t, scale=s, shear=sh, resample=r)
+                out_pil_tensor = torch.from_numpy(np.array(out_pil_img).transpose((2, 0, 1)))
+
                 for fn in [F.affine, scripted_affine]:
                     out_tensor = fn(tensor, angle=a, translate=t, scale=s, shear=sh, resample=r)
-                    out_pil_img = F.affine(pil_img, angle=a, translate=t, scale=s, shear=sh, resample=r)
-                    out_pil_tensor = torch.from_numpy(np.array(out_pil_img).transpose((2, 0, 1)))
-
                     num_diff_pixels = (out_tensor != out_pil_tensor).sum().item() / 3.0
                     ratio_diff_pixels = num_diff_pixels / out_tensor.shape[-1] / out_tensor.shape[-2]
                     # Tolerance : less than 5% of different pixels
@@ -473,6 +473,47 @@ def test_affine(self):
                         )
                     )
 
+    def test_rotate(self):
+        # Tests on square image
+        tensor, pil_img = self._create_data(26, 26)
+        scripted_rotate = torch.jit.script(F.rotate)
+
+        img_size = pil_img.size
+
+        centers = [
+            None,
+            (int(img_size[0] * 0.3), int(img_size[0] * 0.4)),
+            [int(img_size[0] * 0.5), int(img_size[0] * 0.6)]
+        ]
+
+        for r in [0, ]:
+            for a in range(-120, 120, 23):
+                for e in [True, False]:
+                    for c in centers:
+
+                        out_pil_img = F.rotate(pil_img, angle=a, resample=r, expand=e, center=c)
+                        out_pil_tensor = torch.from_numpy(np.array(out_pil_img).transpose((2, 0, 1)))
+                        for fn in [F.rotate, scripted_rotate]:
+                            out_tensor = fn(tensor, angle=a, resample=r, expand=e, center=c)
+
+                            self.assertEqual(
+                                out_tensor.shape,
+                                out_pil_tensor.shape,
+                                msg="{}: {} vs {}".format(
+                                    (r, a, e, c), out_tensor.shape, out_pil_tensor.shape
+                                )
+                            )
+                            num_diff_pixels = (out_tensor != out_pil_tensor).sum().item() / 3.0
+                            ratio_diff_pixels = num_diff_pixels / out_tensor.shape[-1] / out_tensor.shape[-2]
+                            # Tolerance : less than 2% of different pixels
+                            self.assertLess(
+                                ratio_diff_pixels,
+                                0.02,
+                                msg="{}: {}\n{} vs \n{}".format(
+                                    (r, a, e, c), ratio_diff_pixels, out_tensor[0, :7, :7], out_pil_tensor[0, :7, :7]
+                                )
+                            )
+
 
 if __name__ == '__main__':
     unittest.main()

test/test_transforms.py

@@ -1266,7 +1266,7 @@ def test_rotate(self):
         x = np.zeros((100, 100, 3), dtype=np.uint8)
         x[40, 40] = [255, 255, 255]
 
-        with self.assertRaises(TypeError):
+        with self.assertRaisesRegex(TypeError, r"img should be PIL Image"):
             F.rotate(x, 10)
 
         img = F.to_pil_image(x)

torchvision/transforms/functional.py

@@ -756,40 +756,8 @@ def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
     return F_t.adjust_gamma(img, gamma, gain)
 
 
-def rotate(img, angle, resample=False, expand=False, center=None, fill=None):
-    """Rotate the image by angle.
-
-
-    Args:
-        img (PIL Image): PIL Image to be rotated.
-        angle (float or int): In degrees degrees counter clockwise order.
-        resample (``PIL.Image.NEAREST`` or ``PIL.Image.BILINEAR`` or ``PIL.Image.BICUBIC``, optional):
-            An optional resampling filter. See `filters`_ for more information.
-            If omitted, or if the image has mode "1" or "P", it is set to ``PIL.Image.NEAREST``.
-        expand (bool, optional): Optional expansion flag.
-            If true, expands the output image to make it large enough to hold the entire rotated image.
-            If false or omitted, make the output image the same size as the input image.
-            Note that the expand flag assumes rotation around the center and no translation.
-        center (2-tuple, optional): Optional center of rotation.
-            Origin is the upper left corner.
-            Default is the center of the image.
-        fill (n-tuple or int or float): Pixel fill value for area outside the rotated
-            image. If int or float, the value is used for all bands respectively.
-            Defaults to 0 for all bands. This option is only available for ``pillow>=5.2.0``.
-
-    .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters
-
-    """
-    if not F_pil._is_pil_image(img):
-        raise TypeError('img should be PIL Image. Got {}'.format(type(img)))
-
-    opts = _parse_fill(fill, img, '5.2.0')
-
-    return img.rotate(angle, resample, expand, center, **opts)
-
-
 def _get_inverse_affine_matrix(
-        center: List[int], angle: float, translate: List[float], scale: float, shear: List[float]
+        center: List[float], angle: float, translate: List[float], scale: float, shear: List[float]
 ) -> List[float]:
     # Helper method to compute inverse matrix for affine transformation
 
@@ -838,6 +806,56 @@ def _get_inverse_affine_matrix(
     return matrix
 
 
+def rotate(
+        img: Tensor, angle: float, resample: int = 0, expand: bool = False,
+        center: Optional[List[int]] = None, fill: Optional[int] = None
+) -> Tensor:
+    """Rotate the image by angle.
+    The image can be a PIL Image or a Tensor, in which case it is expected
+    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
+
+    Args:
+        img (PIL Image or Tensor): image to be rotated.
+        angle (float or int): rotation angle value in degrees, counter-clockwise.
+        resample (``PIL.Image.NEAREST`` or ``PIL.Image.BILINEAR`` or ``PIL.Image.BICUBIC``, optional):
+            An optional resampling filter. See `filters`_ for more information.
+            If omitted, or if the image has mode "1" or "P", it is set to ``PIL.Image.NEAREST``.
+        expand (bool, optional): Optional expansion flag.
+            If true, expands the output image to make it large enough to hold the entire rotated image.
+            If false or omitted, make the output image the same size as the input image.
+            Note that the expand flag assumes rotation around the center and no translation.
+        center (list or tuple, optional): Optional center of rotation. Origin is the upper left corner.
+            Default is the center of the image.
+        fill (n-tuple or int or float): Pixel fill value for area outside the rotated
+            image. If int or float, the value is used for all bands respectively.
+            Defaults to 0 for all bands. This option is only available for ``pillow>=5.2.0``.
+
+    Returns:
+        PIL Image or Tensor: Rotated image.
+
+    .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters
+
+    """
+    if not isinstance(angle, (int, float)):
+        raise TypeError("Argument angle should be int or float")
+
+    if center is not None and not isinstance(center, (list, tuple)):
+        raise TypeError("Argument center should be a sequence")
+
+    if not isinstance(img, torch.Tensor):
+        return F_pil.rotate(img, angle=angle, resample=resample, expand=expand, center=center, fill=fill)
+
+    center_f = [0.0, 0.0]
+    if center is not None:
+        img_size = _get_image_size(img)
+        # Center is normalized to [-1, +1]
+        center_f = [2.0 * t / s - 1.0 for s, t in zip(img_size, center)]
+    # due to current incoherence of rotation angle direction between affine and rotate implementations
+    # we need to set -angle.
+    matrix = _get_inverse_affine_matrix(center_f, -angle, [0.0, 0.0], 1.0, [0.0, 0.0])
+    return F_t.rotate(img, matrix=matrix, resample=resample, expand=expand, fill=fill)
+
+
 def affine(
         img: Tensor, angle: float, translate: List[int], scale: float, shear: List[float],
         resample: int = 0, fillcolor: Optional[int] = None
@@ -847,7 +865,7 @@ def affine(
     to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
 
     Args:
-        img (PIL Image or Tensor): image to be rotated.
+        img (PIL Image or Tensor): image to transform.
         angle (float or int): rotation angle in degrees between -180 and 180, clockwise direction.
         translate (list or tuple of integers): horizontal and vertical translations (post-rotation translation)
         scale (float): overall scale
@@ -911,7 +929,7 @@ def affine(
     # we need to rescale translate by image size / 2 as its values can be between -1 and 1
     translate = [2.0 * t / s for s, t in zip(img_size, translate)]
 
-    matrix = _get_inverse_affine_matrix([0, 0], angle, translate, scale, shear)
+    matrix = _get_inverse_affine_matrix([0.0, 0.0], angle, translate, scale, shear)
     return F_t.affine(img, matrix=matrix, resample=resample, fillcolor=fillcolor)
 
 

torchvision/transforms/functional_pil.py

@@ -422,3 +422,37 @@ def affine(img, matrix, resample=0, fillcolor=None):
     output_size = img.size
     opts = _parse_fill(fillcolor, img, '5.0.0')
     return img.transform(output_size, Image.AFFINE, matrix, resample, **opts)
+
+
+@torch.jit.unused
+def rotate(img, angle, resample=0, expand=False, center=None, fill=None):
+    """Rotate PIL image by angle.
+
+    Args:
+        img (PIL Image): image to be rotated.
+        angle (float or int): rotation angle value in degrees, counter-clockwise.
+        resample (``PIL.Image.NEAREST`` or ``PIL.Image.BILINEAR`` or ``PIL.Image.BICUBIC``, optional):
+            An optional resampling filter. See `filters`_ for more information.
+            If omitted, or if the image has mode "1" or "P", it is set to ``PIL.Image.NEAREST``.
+        expand (bool, optional): Optional expansion flag.
+            If true, expands the output image to make it large enough to hold the entire rotated image.
+            If false or omitted, make the output image the same size as the input image.
+            Note that the expand flag assumes rotation around the center and no translation.
+        center (2-tuple, optional): Optional center of rotation.
+            Origin is the upper left corner.
+            Default is the center of the image.
+        fill (n-tuple or int or float): Pixel fill value for area outside the rotated
+            image. If int or float, the value is used for all bands respectively.
+            Defaults to 0 for all bands. This option is only available for ``pillow>=5.2.0``.
+
+    Returns:
+        PIL Image: Rotated image.
+
+    .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters
+
+    """
+    if not _is_pil_image(img):
+        raise TypeError("img should be PIL Image. Got {}".format(type(img)))
+
+    opts = _parse_fill(fill, img, '5.2.0')
+    return img.rotate(angle, resample, expand, center, **opts)