|
|
import pytest |
|
|
import torch |
|
|
from torch.autograd import gradcheck |
|
|
|
|
|
import kornia |
|
|
import kornia.testing as utils |
|
|
from kornia.testing import assert_close |
|
|
|
|
|
|
|
|
@pytest.mark.parametrize("batch_size", [0, 1, 5]) |
|
|
@pytest.mark.parametrize("ksize", [3, 11]) |
|
|
@pytest.mark.parametrize("angle", [0.0, 360.0]) |
|
|
@pytest.mark.parametrize("direction", [-1.0, 1.0]) |
|
|
def test_get_motion_kernel2d(batch_size, ksize, angle, direction): |
|
|
if batch_size != 0: |
|
|
angle = torch.tensor([angle] * batch_size) |
|
|
direction = torch.tensor([direction] * batch_size) |
|
|
else: |
|
|
batch_size = 1 |
|
|
kernel = kornia.filters.kernels_geometry.get_motion_kernel2d(ksize, angle, direction) |
|
|
assert kernel.shape == (batch_size, ksize, ksize) |
|
|
assert kernel.sum().item() == pytest.approx(batch_size) |
|
|
|
|
|
|
|
|
class TestMotionBlur: |
|
|
@pytest.mark.parametrize("batch_shape", [(1, 4, 8, 15), (2, 3, 11, 7)]) |
|
|
def test_motion_blur(self, batch_shape, device, dtype): |
|
|
ksize = 5 |
|
|
angle = 200.0 |
|
|
direction = 0.3 |
|
|
|
|
|
input = torch.rand(batch_shape, device=device, dtype=dtype) |
|
|
motion = kornia.filters.MotionBlur(ksize, angle, direction) |
|
|
assert motion(input).shape == batch_shape |
|
|
|
|
|
def test_noncontiguous(self, device, dtype): |
|
|
batch_size = 3 |
|
|
inp = torch.rand(3, 5, 5, device=device, dtype=dtype).expand(batch_size, -1, -1, -1) |
|
|
|
|
|
kernel_size = 3 |
|
|
angle = 200.0 |
|
|
direction = 0.3 |
|
|
actual = kornia.filters.motion_blur(inp, kernel_size, angle, direction) |
|
|
assert_close(actual, actual) |
|
|
|
|
|
def test_gradcheck(self, device, dtype): |
|
|
batch_shape = (1, 3, 4, 5) |
|
|
ksize = 9 |
|
|
angle = 34.0 |
|
|
direction = -0.2 |
|
|
|
|
|
input = torch.rand(batch_shape, device=device, dtype=dtype) |
|
|
input = utils.tensor_to_gradcheck_var(input) |
|
|
assert gradcheck( |
|
|
kornia.filters.motion_blur, (input, ksize, angle, direction, "replicate"), raise_exception=True |
|
|
) |
|
|
|
|
|
@pytest.mark.skip("angle can be Union") |
|
|
def test_jit(self, device, dtype): |
|
|
img = torch.rand(2, 3, 4, 5, device=device, dtype=dtype) |
|
|
ksize = 5 |
|
|
angle = 65.0 |
|
|
direction = 0.1 |
|
|
op = kornia.filters.motion_blur |
|
|
op_script = torch.jit.script(op) |
|
|
actual = op_script(img, ksize, angle, direction) |
|
|
expected = op(img, ksize, angle, direction) |
|
|
assert_close(actual, expected) |
|
|
|
