Expand the channels to 3 if the user requested as such

test/test_transforms_v2.py

@@ -4935,9 +4935,10 @@ def test_transform(self, transform, make_input):
         check_transform(transform, make_input())
 
     @pytest.mark.parametrize("num_output_channels", [1, 3])
+    @pytest.mark.parametrize("color_space", ["RGB", "GRAY"])
     @pytest.mark.parametrize("fn", [F.rgb_to_grayscale, transform_cls_to_functional(transforms.Grayscale)])
-    def test_image_correctness(self, num_output_channels, fn):
-        image = make_image(dtype=torch.uint8, device="cpu")
+    def test_image_correctness(self, num_output_channels, color_space, fn):
+        image = make_image(dtype=torch.uint8, device="cpu", color_space=color_space)
 
         actual = fn(image, num_output_channels=num_output_channels)
         expected = F.to_image(F.rgb_to_grayscale(F.to_pil_image(image), num_output_channels=num_output_channels))

torchvision/transforms/v2/functional/_color.py

@@ -33,9 +33,13 @@ def rgb_to_grayscale(inpt: torch.Tensor, num_output_channels: int = 1) -> torch.
 def _rgb_to_grayscale_image(
     image: torch.Tensor, num_output_channels: int = 1, preserve_dtype: bool = True
 ) -> torch.Tensor:
-    if image.shape[-3] == 1:
+    # TODO: Maybe move the validation that num_output_channels is 1 or 3 to this function instead of callers.
+    if image.shape[-3] == 1 and num_output_channels == 1:
         return image.clone()
-
+    if image.shape[-3] == 1 and num_output_channels == 3:
+        s = [-1] * len(image.shape)
+        s[-3] = 3
+        return image.expand(s)
     r, g, b = image.unbind(dim=-3)
     l_img = r.mul(0.2989).add_(g, alpha=0.587).add_(b, alpha=0.114)
     l_img = l_img.unsqueeze(dim=-3)