def test_shape_tuple(self, device, dtype):
inp = torch.zeros(3, 2, 41, 41, device=device, dtype=dtype)
SP = kornia.geometry.ScalePyramid(n_levels=1, min_size=30)
out = SP(inp)
assert len(out) == 3
assert len(out[0]) == 1
assert len(out[1]) == 1
assert len(out[2]) == 1
def test_shape_tuple(self):
inp = torch.zeros(3, 2, 6, 6)
SP = kornia.geometry.ScalePyramid(n_levels=1)
out = SP(inp)
assert len(out) == 3
assert len(out[0]) == 1
assert len(out[1]) == 1
assert len(out[2]) == 1
def test_blur_order(self, device, dtype):
inp = torch.rand(1, 1, 31, 31, device=device, dtype=dtype)
SP = kornia.geometry.ScalePyramid(n_levels=3)
sp, _, _ = SP(inp)
for _, pyr_level in enumerate(sp):
for _, img in enumerate(pyr_level):
img = img.squeeze().view(3, -1)
max_per_blur_level_val, _ = img.max(dim=1)
assert torch.argmax(max_per_blur_level_val).item() == 0
def test_blur_order(self):
inp = torch.rand(1, 1, 12, 12)
SP = kornia.geometry.ScalePyramid(n_levels=5)
sp, sigmas, pd = SP(inp)
for i, pyr_level in enumerate(sp):
for ii, img in enumerate(pyr_level):
img = img.squeeze().view(5, -1)
max_per_blur_level_val, _ = img.max(dim=1)
assert torch.argmax(max_per_blur_level_val).item() == 0
return
def test_symmetry_preserving(self):
inp = torch.zeros(1, 1, 12, 12)
inp[0, 0, 4:8, 4:8] = 1.0
SP = kornia.geometry.ScalePyramid(n_levels=5)
sp, sigmas, pd = SP(inp)
for i, pyr_level in enumerate(sp):
for ii, img in enumerate(pyr_level):
img = img.squeeze()
assert torch.allclose(img, img.flip(1))
assert torch.allclose(img, img.flip(2))
return
def test_symmetry_preserving(self, device, dtype):
PS = 16
R = 2
inp = torch.zeros(1, 1, PS, PS, device=device, dtype=dtype)
inp[..., PS // 2 - R:PS // 2 + R, PS // 2 - R:PS // 2 + R] = 1.0
SP = kornia.geometry.ScalePyramid(n_levels=3)
sp, _, _ = SP(inp)
for _, pyr_level in enumerate(sp):
for _, img in enumerate(pyr_level):
img = img.squeeze()
assert_close(img, img.flip(1))
assert_close(img, img.flip(2))
def test_symmetry_preserving(self, device, dtype):
PS = 16
R = 2
inp = torch.zeros(1, 1, PS, PS, device=device, dtype=dtype)
inp[..., PS // 2 - R:PS // 2 + R, PS // 2 - R:PS // 2 + R] = 1.0
SP = kornia.geometry.ScalePyramid(n_levels=3)
sp, sigmas, pd = SP(inp)
for i, pyr_level in enumerate(sp):
for ii, img in enumerate(pyr_level):
img = img.squeeze()
assert torch.allclose(img, img.flip(1), atol=1e-5, rtol=1e-3)
assert torch.allclose(img, img.flip(2), atol=1e-5, rtol=1e-3)
return
def test_n_levels_shape(self, device, dtype):
inp = torch.zeros(1, 1, 32, 32, device=device, dtype=dtype)
SP = kornia.geometry.ScalePyramid(n_levels=3)
sp, _, _ = SP(inp)
assert sp[0].shape == (1, 1, 3 + 3, 32, 32)
def test_shape_batch_double(self, device, dtype):
inp = torch.zeros(3, 2, 31, 31, device=device, dtype=dtype)
SP = kornia.geometry.ScalePyramid(n_levels=1, double_image=True)
sp, _, _ = SP(inp)
assert sp[0].shape == (3, 2, 1 + 3, 62, 62)
def test_shape_batch(self, device, dtype):
inp = torch.zeros(3, 2, 31, 31, device=device, dtype=dtype)
SP = kornia.geometry.ScalePyramid(n_levels=1)
sp, _, _ = SP(inp)
assert sp[0].shape == (3, 2, 3 + 1, 31, 31)
def sp_tuple(img):
sp, _, _ = SP()(img)
return tuple(sp)
def test_n_levels_shape(self):
inp = torch.zeros(1, 1, 6, 6)
SP = kornia.geometry.ScalePyramid(n_levels=5)
sp, sigmas, pd = SP(inp)
assert sp[0].shape == (1, 5, 1, 6, 6)
def test_shape_batch(self):
inp = torch.zeros(3, 2, 6, 6)
SP = kornia.geometry.ScalePyramid(n_levels=1)
sp, sigmas, pd = SP(inp)
assert sp[0].shape == (3, 1, 2, 6, 6)
def sp_tuple(img):
sp, sigmas, pd = SP()(img)
return tuple(sp)
def test_shape_batch_double(self, device):
inp = torch.zeros(3, 2, 6, 6).to(device)
SP = kornia.geometry.ScalePyramid(n_levels=1, double_image=True)
sp, sigmas, pd = SP(inp)
assert sp[0].shape == (3, 1, 2, 12, 12)
def test_n_levels_shape(self, device):
inp = torch.zeros(1, 1, 32, 32).to(device)
SP = kornia.geometry.ScalePyramid(n_levels=3)
sp, sigmas, pd = SP(inp)
assert sp[0].shape == (1, 3, 1, 32, 32)