def yuv_to_rgb(x: torch.Tensor) -> torch.Tensor:
transformation_matrix = torch.tensor(
((1.000, 0.000, 1.140), (1.000, -0.395, -0.581),
(1.000, 2.032, 0.000)),
**tensor_meta(x),
)
return transform_channels_linearly(x, transformation_matrix)
def rgb_to_yuv(x: torch.Tensor) -> torch.Tensor:
transformation_matrix = torch.tensor(
((0.299, 0.587, 0.114), (-0.147, -0.289, 0.436),
(0.615, -0.515, -0.100)),
**tensor_meta(x),
)
return transform_channels_linearly(x, transformation_matrix)
def test_tensor_meta_kwargs(self):
dtype = torch.bool
x = torch.empty(())
actual = meta.tensor_meta(x, dtype=dtype)["dtype"]
desired = dtype
self.assertEqual(actual, desired)
def test_tensor_meta(self):
tensor_meta = {"dtype": torch.bool, "device": torch.device("cpu")}
x = torch.empty((), **tensor_meta)
actual = meta.tensor_meta(x)
desired = tensor_meta
self.assertDictEqual(actual, desired)
def test_kwargs(main):
dtype = torch.bool
x = torch.empty(())
actual = meta.tensor_meta(x, dtype=dtype)["dtype"]
desired = dtype
assert actual == desired
def test_main(self):
tensor_meta = {"dtype": torch.bool, "device": torch.device("cpu")}
x = torch.empty((), **tensor_meta)
actual = meta.tensor_meta(x)
desired = tensor_meta
assert actual == desired
def rgb_to_grayscale(x: torch.Tensor) -> torch.Tensor:
transformation_matrix = torch.tensor(((0.299, 0.587, 0.114), ),
**tensor_meta(x))
return transform_channels_linearly(x, transformation_matrix)
def forward(self, input: torch.Tensor) -> torch.Tensor:
size = self._extract_size(input)
meta = meta_.tensor_meta(input)
noise = torch.rand(size, **meta)
return join_channelwise(input, noise)