def test_lpips_loss_raises_if_wrong_reduction(x, y) -> None:
for mode in ['mean', 'sum', 'none']:
LPIPS(reduction=mode)(x, y)
for mode in [None, 'n', 2]:
with pytest.raises(ValueError):
LPIPS(reduction=mode)(x, y)
def test_lpips_loss_raises_if_wrong_reduction(prediction: torch.Tensor,
target: torch.Tensor) -> None:
for mode in ['mean', 'sum', 'none']:
LPIPS(reduction=mode)(prediction, target)
for mode in [None, 'n', 2]:
with pytest.raises(KeyError):
LPIPS(reduction=mode)(prediction, target)
def test_lpips_loss_forward_for_special_cases(x, y, expectation: Any,
value: float) -> None:
loss = LPIPS()
with expectation:
loss_value = loss(x, y)
assert torch.isclose(loss_value, torch.tensor(value), atol=1e-6), \
f'Expected loss value to be equal to target value. Got {loss_value} and {value}'
def test_lpips_computes_grad(x, y, device: str) -> None:
x.requires_grad_()
loss_value = LPIPS()(x.to(device), y.to(device))
loss_value.backward()
assert x.grad is not None, NONE_GRAD_ERR_MSG
def test_lpips_loss_forward(input_tensors: Tuple[torch.Tensor, torch.Tensor],
device: str) -> None:
x, y = input_tensors
loss = LPIPS()
loss(x.to(device), y.to(device))
def test_lpips_loss_init() -> None:
LPIPS()
def test_lpips_computes_grad(prediction: torch.Tensor, target: torch.Tensor,
device: str) -> None:
prediction.requires_grad_()
loss_value = LPIPS()(prediction.to(device), target.to(device))
loss_value.backward()
assert prediction.grad is not None, NONE_GRAD_ERR_MSG
def test_lpips_loss_forward(input_tensors: Tuple[torch.Tensor, torch.Tensor],
device: str) -> None:
prediction, target = input_tensors
loss = LPIPS()
loss(prediction.to(device), target.to(device))
