def test_scattering2d_errors():
S = Scattering2D(3, (32, 32))
if backend.NAME == 'skcuda':
S.cuda()
with pytest.raises(TypeError) as record:
S(None)
assert ('input should be' in record.value.args[0])
x = torch.randn(4, 4)
y = x[::2, ::2]
with pytest.raises(RuntimeError) as record:
S(y)
assert ('must be contiguous' in record.value.args[0])
x = torch.randn(31, 31)
with pytest.raises(RuntimeError) as record:
S(x)
assert ('Tensor must be of spatial size' in record.value.args[0])
S = Scattering2D(3, (32, 32), pre_pad=True)
with pytest.raises(RuntimeError) as record:
S(x)
assert ('Padded tensor must be of spatial size' in record.value.args[0])
def test_Scattering2D():
test_data_dir = os.path.dirname(__file__)
data = torch.load(os.path.join(test_data_dir, 'test_data_2d.pt'))
x = data['x']
S = data['Sx']
J = data['J']
# we need to reorder S from interleaved (how it's saved) to o0, o1, o2
# (which is how it's now computed)
o0, o1, o2 = reorder_coefficients_from_interleaved(J, L=8)
reorder = torch.from_numpy(np.concatenate((o0, o1, o2)))
S = S[..., reorder, :, :]
pre_pad = data['pre_pad']
M = x.shape[2]
N = x.shape[3]
import kymatio.scattering2d.backend as backend
if backend.NAME == 'skcuda':
print('skcuda backend tested!')
# First, let's check the Jit
scattering = Scattering2D(J, shape=(M, N), pre_pad=pre_pad)
scattering.cuda()
x = x.cuda()
S = S.cuda()
y = scattering(x)
assert ((S - y)).abs().max() < 1e-6
elif backend.NAME == 'torch':
# Then, let's check when using pure pytorch code
scattering = Scattering2D(J, shape=(M, N), pre_pad=pre_pad)
Sg = []
for device in devices:
if device == 'gpu':
print('torch-gpu backend tested!')
x = x.cuda()
scattering.cuda()
S = S.cuda()
Sg = scattering(x)
else:
print('torch-cpu backend tested!')
x = x.cpu()
S = S.cpu()
scattering.cpu()
Sg = scattering(x)
assert (Sg - S).abs().max() < 1e-6
def test_input_size_agnostic():
for N in [31, 32, 33]:
for J in [2, 4]:
scattering = Scattering2D(J, shape=(N, N))
x = torch.zeros(3, 3, N, N)
if backend.NAME == 'skcuda':
x = x.cuda()
scattering.cuda()
S = scattering(x)
scattering = Scattering2D(J, shape=(N, N), pre_pad=True)
x = torch.zeros(3, 3, scattering.M_padded, scattering.N_padded)
if backend.NAME == 'skcuda':
x = x.cuda()
scattering.cuda()
S = scattering(x)
def test_input_size_agnostic():
for N in [31, 32, 33]:
for J in [1, 2, 4]:
scattering = Scattering2D(J, shape=(N, N))
x = torch.zeros(3, 3, N, N)
if backend.NAME == 'skcuda':
x = x.cuda()
scattering.cuda()
S = scattering(x)
scattering = Scattering2D(J, shape=(N, N), pre_pad=True)
x = torch.zeros(3, 3, scattering.M_padded, scattering.N_padded)
if backend.NAME == 'skcuda':
x = x.cuda()
scattering.cuda()
S = scattering(x)
N = 32
J = 5
scattering = Scattering2D(J, shape=(N, N))
x = torch.zeros(3, 3, N, N)
if backend.NAME == 'skcuda':
x = x.cuda()
scattering.cuda()
S = scattering(x)
assert (S.shape[-2:] == (1, 1))
N = 32
J = 5
scattering = Scattering2D(J, shape=(N + 5, N))
x = torch.zeros(3, 3, N + 5, N)
if backend.NAME == 'skcuda':
x = x.cuda()
scattering.cuda()
S = scattering(x)
assert (S.shape[-2:] == (1, 1))
def test_Scattering2D():
test_data_dir = os.path.dirname(__file__)
data = torch.load(os.path.join(test_data_dir, 'test_data_2d.pt'))
x = data['x']
S = data['Sx']
J = data['J']
pre_pad = data['pre_pad']
M = x.shape[2]
N = x.shape[3]
import kymatio.scattering2d.backend as backend
if backend.NAME == 'skcuda':
print('skcuda backend tested!')
# First, let's check the Jit
scattering = Scattering2D(M, N, J, pre_pad=pre_pad)
scattering.cuda()
x = x.cuda()
S = S.cuda()
y = scattering(x)
assert ((S - y)).abs().max() < 1e-6
elif backend.NAME == 'torch':
# Then, let's check when using pure pytorch code
scattering = Scattering2D(M, N, J, pre_pad=pre_pad)
Sg = []
for device in devices:
if device == 'gpu':
print('torch-gpu backend tested!')
x = x.cuda()
scattering.cuda()
S = S.cuda()
Sg = scattering(x)
else:
print('torch-cpu backend tested!')
x = x.cpu()
S = S.cpu()
scattering.cpu()
Sg = scattering(x)
assert (Sg - S).abs().max() < 1e-6
def test_Scattering2D():
test_data_dir = os.path.dirname(__file__)
data = torch.load(os.path.join(test_data_dir, 'test_data_2d.pt'),
map_location='cpu')
x = data['x'].view(7, 3, 128, 128)
S = data['S'].view(7, 3, 417, 8, 8)
import kymatio.scattering2d.backend as backend
if backend.NAME == 'skcuda':
print('skcuda backend tested!')
# First, let's check the Jit
scattering = Scattering2D(128, 128, 4, pre_pad=False)
scattering.cuda()
x = x.cuda()
S = S.cuda()
y = scattering(x)
assert ((S - y)).abs().max() < 1e-6
elif backend.NAME == 'torch':
# Then, let's check when using pure pytorch code
scattering = Scattering2D(128, 128, 4, pre_pad=False)
Sg = []
for device in devices:
if device == 'gpu':
print('torch-gpu backend tested!')
x = x.cuda()
scattering.cuda()
S = S.cuda()
Sg = scattering(x)
else:
print('torch-cpu backend tested!')
x = x.cpu()
S = S.cpu()
scattering.cpu()
Sg = scattering(x)
assert (Sg - S).abs().max() < 1e-6
def test_batch_shape_agnostic():
J = 3
L = 8
shape = (32, 32)
shape_ds = tuple(n // 2**J for n in shape)
S = Scattering2D(J, shape, L)
with pytest.raises(RuntimeError) as ve:
S(torch.zeros(()))
assert "at least two" in ve.value.args[0]
with pytest.raises(RuntimeError) as ve:
S(torch.zeros((32, )))
assert "at least two" in ve.value.args[0]
x = torch.zeros(shape)
if backend.NAME == 'skcuda':
x = x.cuda()
S.cuda()
Sx = S(x)
assert len(Sx.shape) == 3
assert Sx.shape[-2:] == shape_ds
n_coeffs = Sx.shape[-3]
test_shapes = ((1, ) + shape, (2, ) + shape, (2, 2) + shape,
(2, 2, 2) + shape)
for test_shape in test_shapes:
x = torch.zeros(test_shape)
if backend.NAME == 'skcuda':
x = x.cuda()
Sx = S(x)
assert len(Sx.shape) == len(test_shape) + 1
assert Sx.shape[-2:] == shape_ds
assert Sx.shape[-3] == n_coeffs
assert Sx.shape[:-3] == test_shape[:-2]