def test_cpu_approx(self, test_case_description, sigmas, input, expected):
# Params to determine the implementation to test
device = torch.device("cpu")
fast_approx = True
# Create input tensor and apply filter
input_tensor = torch.from_numpy(np.array(input)).to(dtype=torch.float, device=device)
output = BilateralFilter.apply(input_tensor, *sigmas, fast_approx).cpu().numpy()
# Ensure result are as expected
np.testing.assert_allclose(output, expected, atol=1e-5)
def test_cuda_precise(self, test_case_description, sigmas, input, expected):
# Skip this test
if not torch.cuda.is_available():
return
# Params to determine the implementation to test
device = torch.device("cuda")
fast_approx = False
# Create input tensor and apply filter
input_tensor = torch.from_numpy(np.array(input)).to(dtype=torch.float, device=device)
output = BilateralFilter.apply(input_tensor, *sigmas, fast_approx).cpu().numpy()
# Ensure result are as expected
np.testing.assert_allclose(output, expected, atol=1e-5)