def test_multi_scale_loss_gmsd_modes(x, y, device: str) -> None:
for reduction in ['mean', 'sum', 'none']:
MultiScaleGMSDLoss(reduction=reduction)(x.to(device), y.to(device))
for reduction in ['DEADBEEF', 'random']:
with pytest.raises(ValueError):
MultiScaleGMSDLoss(reduction=reduction)(x.to(device), y.to(device))
def test_multi_scale_loss_gmsd_modes(prediction: torch.Tensor, target: torch.Tensor, device: str) -> None:
for reduction in ['mean', 'sum', 'none']:
MultiScaleGMSDLoss(reduction=reduction)(prediction.to(device), target.to(device))
for reduction in ['DEADBEEF', 'random']:
with pytest.raises(KeyError):
MultiScaleGMSDLoss(reduction=reduction)(prediction.to(device), target.to(device))
def test_multi_scale_gmsd_loss_forward_backward(prediction: torch.Tensor,
target: torch.Tensor,
device: str) -> None:
prediction.requires_grad_()
loss_value = MultiScaleGMSDLoss(chromatic=True)(prediction.to(device),
target.to(device))
loss_value.backward()
assert torch.isfinite(prediction.grad).all(), LEAF_VARIABLE_ERROR_MESSAGE
def test_multi_scale_gmsd_loss_supports_different_data_ranges(x, y, device: str) -> None:
x_255 = x * 255
y_255 = y * 255
loss = MultiScaleGMSDLoss()
measure = loss(x.to(device), y.to(device))
loss_255 = MultiScaleGMSDLoss(data_range=255)
measure_255 = loss_255(x_255.to(device), y_255.to(device))
diff = torch.abs(measure_255 - measure)
assert diff <= 1e-4, f'Result for same tensor with different data_range should be the same, got {diff}'
def test_multi_scale_gmsd_loss_supports_different_data_ranges(prediction: torch.Tensor, target: torch.Tensor,
device: str) -> None:
prediction_255 = prediction * 255
target_255 = target * 255
loss = MultiScaleGMSDLoss()
measure = loss(prediction.to(device), target.to(device))
loss_255 = MultiScaleGMSDLoss(data_range=255)
measure_255 = loss_255(prediction_255.to(device), target_255.to(device))
diff = torch.abs(measure_255 - measure)
assert diff <= 1e-4, f'Result for same tensor with different data_range should be the same, got {diff}'
def test_multi_scale_gmsd_loss_raise_exception_for_small_images(
device: str) -> None:
target = torch.ones(3, 1, 32, 32)
prediction = torch.zeros(3, 1, 32, 32)
loss = MultiScaleGMSDLoss(scale_weights=[3., 4., 2., 1., 1.])
with pytest.raises(ValueError):
loss(prediction.to(device), target.to(device))
def test_multi_scale_gmsd_loss_fails_for_greyscale_tensors_chromatic_flag(
device: str) -> None:
loss = MultiScaleGMSDLoss(chromatic=True)
target = torch.ones(2, 1, 96, 96)
prediction = torch.zeros(2, 1, 96, 96)
with pytest.raises(AssertionError):
loss(prediction.to(device), target.to(device))
def test_multi_scale_gmsd_loss_zero_for_equal_tensors(prediction: torch.Tensor,
device: str) -> None:
loss = MultiScaleGMSDLoss()
target = prediction.clone()
measure = loss(prediction.to(device), target.to(device))
assert measure.abs(
) <= 1e-6, f'MultiScaleGMSD for equal tensors must be 0, got {measure}'
def test_multi_scale_gmsd_loss_raise_exception_for_small_images(
device: str) -> None:
y = torch.ones(3, 1, 32, 32)
x = torch.zeros(3, 1, 32, 32)
loss = MultiScaleGMSDLoss(scale_weights=torch.tensor([3., 4., 2., 1., 2.]))
with pytest.raises(ValueError):
loss(x.to(device), y.to(device))
def test_multi_scale_gmsd_loss_supports_custom_weights(prediction: torch.Tensor, target: torch.Tensor,
device: str) -> None:
loss = MultiScaleGMSDLoss(scale_weights=[3., 4., 2., 1., 2.])
loss(prediction.to(device), target.to(device))
loss = MultiScaleGMSDLoss(scale_weights=torch.tensor([3., 4., 2., 1., 2.]))
loss(prediction.to(device), target.to(device))
def test_multi_scale_gmsd_loss_supports_greyscale_tensors(device: str) -> None:
loss = MultiScaleGMSDLoss()
target = torch.ones(2, 1, 96, 96)
prediction = torch.zeros(2, 1, 96, 96)
loss(prediction.to(device), target.to(device))
def test_multi_scale_gmsd_loss_supports_custom_weights(x, y, device: str) -> None:
loss = MultiScaleGMSDLoss(scale_weights=torch.tensor([3., 4., 2., 1., 2.]))
loss(x.to(device), y.to(device))
def test_multi_scale_gmsd_loss_supports_greyscale_tensors(device: str) -> None:
loss = MultiScaleGMSDLoss()
y = torch.ones(2, 1, 96, 96)
x = torch.zeros(2, 1, 96, 96)
loss(x.to(device), y.to(device))
def test_multi_scale_gmsd_loss_zero_for_equal_tensors(x, device: str) -> None:
loss = MultiScaleGMSDLoss()
y = x.clone()
measure = loss(x.to(device), y.to(device))
assert measure.abs() <= 1e-6, f'MultiScaleGMSD for equal tensors must be 0, got {measure}'
def test_multi_scale_gmsd_loss_forward_backward(x, y, device: str) -> None:
x.requires_grad_()
loss_value = MultiScaleGMSDLoss(chromatic=True)(x.to(device), y.to(device))
loss_value.backward()
assert torch.isfinite(x.grad).all(), LEAF_VARIABLE_ERROR_MESSAGE
