def test_patches3d_stride():
vol = np.random.rand(20, 20, 20)
size = (10, 10, 10)
patches = preprocess.Patches(vol, size, (1, 1, 11))
with pytest.raises(ValueError):
patches.reconstruct(patches.patches)
patches = preprocess.Patches(vol, size, (1, 1, 10))
vol_a = patches.reconstruct(patches.patches)
patch_size = 10 * 10 * 10
n_patches = 11 * 11 * 2
assert patches.shape == (patch_size, n_patches)
assert np.allclose(vol, vol_a)
patches = preprocess.Patches(vol, size, (2, 2, 2))
recon = patches.reconstruct(patches.patches)
assert patches.shape == (10 * 10 * 10, 6 * 6 * 6)
assert np.allclose(vol, recon)
with pytest.raises(ValueError):
too_many_dims = np.zeros((2, 2, 2, 2))
preprocess.Patches(too_many_dims, [1] * 3, [1] * 3)
with pytest.raises(ValueError):
too_few_dims = np.zeros((2, 2))
preprocess.Patches(too_few_dims, [1] * 3, [1] * 3)
def test_patches3d_check_size():
vol = np.zeros((25, 25, 25), dtype=np.float64)
patches = preprocess.Patches(vol, (10, 10, 10), (1, 1, 1))
size, dims = patches._check_size(False, True)
size = [size] + dims
assert size[0] == (25 - 10 + 1)**3
assert size[1] == 25 - 10 + 1
assert size[2] == 25 - 10 + 1
assert size[3] == 25 - 10 + 1
with pytest.raises(MemoryError):
p = preprocess.Patches(np.zeros((256, 256, 256), dtype=np.float64),
(10, 10, 10), (2, 2, 2))
_ = p.patches
p = preprocess.Patches(vol, (10, 10, 10), (2, 2, 2))
size, dims = p._check_size(False, True)
x, y, z = dims
exp = (25 - 10 + 1) // 2 + 1
assert x == exp
assert y == exp
assert z == exp
assert size == exp**3 and size == x * y * z
with pytest.raises(ValueError):
preprocess.Patches(np.zeros((1, 1, 1, 1)), (1, 1, 1), (1, 1, 1))
def test_patches2d_bad_input():
with pytest.raises(ValueError):
preprocess.Patches(np.zeros((1, 1, 1, 1)), 10)
with pytest.raises(ValueError):
preprocess.Patches(np.arange(2), 2)
with pytest.raises(ValueError):
preprocess.Patches(np.zeros((10, 10)), 10, max_patches=10)
def test_patches3d_static():
vol = np.random.rand(20, 20, 20)
patches = preprocess.Patches(vol, (10, 10, 10), (1, 1, 1))
assert patches.shape == (10 * 10 * 10, 11**3), '3D patches incorrect shape'
vol_a = patches.reconstruct(patches.patches)
assert np.allclose(vol, vol_a), 'Recon 3D patches incorr values'
with pytest.raises(MemoryError):
vol = np.random.rand(256, 256, 256)
preprocess.Patches(vol, (12, 12, 12), (1, 1, 1)).patches
def test_patches_mean():
img = np.random.rand(100, 100)
patches = preprocess.Patches(img.copy(), 10)
patches.remove_mean(add_back=True)
new = patches.reconstruct(patches.patches)
assert np.allclose(img, new)
patches = preprocess.Patches(img.copy(), 10)
patches.remove_mean(add_back=False)
new = patches.reconstruct(patches.patches)
assert not np.allclose(img - img.mean(), new)
def test_patches_3d():
vol = np.ones((10, 10, 10))
patches = preprocess.Patches(vol, 8)
assert patches.shape == (8 * 8 * 8, (10 - 8 + 1)**3)
assert patches.size == 8
new = patches.reconstruct(patches.patches)
assert np.array_equal(vol, new)
def sparse_signal():
dir_path = os.path.dirname(os.path.realpath(__file__))
img = np.load(os.path.join(dir_path, 'test_img1.npy'))
img = img.astype(float)
img = filters.threshold(utils.normalize(img), 0.4)
return preprocess.Patches(img, 8).patches
def test_patches_reconstruct():
img = np.random.rand(128, 128)
with pytest.raises(ValueError):
patches = preprocess.Patches(img.copy(), 8, order='F')
patches.reconstruct(patches.patches)
patches = preprocess.Patches(img.copy(), 8)
before = patches.patches.copy()
recon = patches.reconstruct(patches.patches)
assert recon.shape == img.shape
assert np.allclose(img, recon)
assert np.array_equal(before, patches.patches)
new_patches = patches.patches + 10
patches.reconstruct(new_patches, save=True)
assert np.array_equal(new_patches, patches.patches)
assert not np.allclose(before, patches.patches)
def test_patches3d_batch():
vol = np.random.rand(20, 20, 20)
for i in range(9):
vol[:, i, :] = i
size = (9, 9, 9)
stride = (1, 1, 1)
corr = preprocess.Patches(vol.copy(), size, stride)
corr = corr.patches
iter_n_patches = 0
patches = preprocess.Patches(vol, size, stride)
iter_patches = None
for batch in patches.generator(11):
if iter_patches is None:
iter_patches = batch.copy()
else:
iter_patches = np.hstack((iter_patches, batch.copy()))
iter_n_patches += batch.shape[1]
assert iter_n_patches == corr.shape[1]
assert corr.shape == iter_patches.shape
assert np.array_equal(corr, iter_patches)
assert patches.n_patches == (20 - 9 + 1)**3
stride = (1, 1, 3)
patches = preprocess.Patches(vol, size, stride)
assert patches.n_patches == (20 - 9 + 1)**2 * (
(20 - 9 + 1) // stride[2] + 1)
stride = (1, 3, 1)
patches = preprocess.Patches(vol, size, stride)
assert patches.n_patches == (20 - 9 + 1)**2 * (
(20 - 9 + 1) // stride[1] + 1)
stride = (3, 1, 1)
patches = preprocess.Patches(vol, size, stride)
assert patches.n_patches == (20 - 9 + 1)**2 * (
(20 - 9 + 1) // stride[0] + 1)
def test_patches3d_reconstruct_batch():
vol = np.random.rand(10, 10, 10)
size = (7, 7, 7)
stride = (1, 1, 1)
patches = preprocess.Patches(vol, size, stride)
for batch, reconstruct in patches.generator(10, callback=True):
reconstruct(batch)
assert np.array_equal(patches.image, vol)
vol = np.random.rand(20, 20, 20)
size = (7, 7, 7)
stride = (3, 1, 2)
patches = preprocess.Patches(vol, size, stride)
for batch, reconstruct in patches.generator(10, callback=True):
reconstruct(batch)
assert np.array_equal(patches.image, vol)
patches = preprocess.Patches(np.zeros((256, 256, 256)), [10] * 3, [1] * 3)
with pytest.raises(MemoryError):
patches.generator(1e11)
with pytest.raises(MemoryError):
patches.generator(1e10)
vol = np.random.rand(20, 20, 20)
size = (7, 7, 7)
stride = (1, 1, 7)
patches = preprocess.Patches(vol, size, stride)
for batch, recon in patches.generator(10, callback=True):
recon(batch)
assert np.allclose(vol, patches.image)
def test_patches():
img = helpers.get_image(1)
patch_size = 8
patch_maker = preprocess.Patches(img, patch_size)
patches = patch_maker.patches
reconstructed = patch_maker.reconstruct(patches)
assert patches.shape[0] == patch_size**2
assert np.allclose(img, reconstructed)
patch_size = 16
patch_maker = preprocess.Patches(img, patch_size)
assert patch_maker.size == patch_size
patches = patch_maker.patches
reconstructed = patch_maker.reconstruct(patches)
assert patches.shape[0] == patch_size**2
assert np.allclose(img, reconstructed)
img = np.random.rand(256, 256, 256)
patch_size = 16
with pytest.raises(MemoryError):
preprocess.Patches(img, patch_size).patches
def test_expand_index():
vol = np.random.rand(15, 15, 15)
stride = (1, 1, 1)
size = (10, 10, 10)
patches = preprocess.Patches(vol, size, stride)
count = 0
n_patches, dims = patches._check_size(False, True)
xx, yy, zz = dims
for i in range(0, xx, stride[0]):
for j in range(0, yy, stride[1]):
for k in range(0, zz, stride[2]):
ii, jj, kk = patches._expand_index(count)
msg = 'stride={}, {} -> {} != {}'.format(
stride, count, [ii, jj, kk], [i, j, k])
assert i == ii, msg
assert j == jj, msg
assert k == kk, msg
count += 1
def f(img_num):
dir_path = os.path.dirname(os.path.realpath(__file__))
img = np.load(os.path.join(dir_path, 'test_img%d.npy' % img_num))
img = img.astype(float)
return preprocess.Patches(img, 8).patches