def test_loss_same_images(self, size):
"""Tests that passing same images results in a loss close to 0.0."""
tensor_shape = np.random.randint(size, 6, size=3).tolist()
image = np.random.uniform(0.0, 1.0, size=tensor_shape + [1])
laplacian, _ = matting.build_matrices(image, size=size)
loss = matting.loss(image, laplacian)
self.assertAllClose(loss, 0.0, atol=1e-4)
def test_loss_same_images(self):
"""Tests that passing same images results in a loss close to 0.0."""
shape = np.random.randint(3, 5, size=3).tolist()
image = np.random.uniform(0.0, 1.0, size=shape + [1])
weights = matting.laplacian_weights(image)
loss = matting.loss(image, weights)
self.assertAllClose(loss, 0.0, atol=1e-4)
def test_loss_positive(self):
"""Tests that the loss is always greater or equal to 0.0."""
shape = np.random.randint(3, 5, size=3).tolist()
image = tf.random.uniform(minval=0.0, maxval=1.0, shape=shape + [3])
matte = tf.random.uniform(minval=0.0, maxval=1.0, shape=shape + [1])
weights = matting.laplacian_weights(image)
loss = matting.loss(matte, weights)
self.assertAllGreaterEqual(loss, 0.0)
def test_loss_positive(self, size):
"""Tests that the loss is always greater or equal to 0.0."""
tensor_shape = np.random.randint(size, 6, size=3).tolist()
image = tf.random.uniform(minval=0.0, maxval=1.0, shape=tensor_shape + [3])
matte = tf.random.uniform(minval=0.0, maxval=1.0, shape=tensor_shape + [1])
laplacian, _ = matting.build_matrices(image, size=size)
loss = matting.loss(matte, laplacian)
self.assertAllGreaterEqual(loss, 0.0)
def test_loss_jacobian_random(self, size):
"""Tests the Jacobian of the matting loss function."""
tensor_shape = np.random.randint(size, 6, size=3)
matte_init = np.random.uniform(0.0, 1.0, size=tensor_shape.tolist() + [1])
tensor_shape[1:3] -= (size - 1)
laplacian_init = np.random.uniform(
0.0, 1.0, size=tensor_shape.tolist() + [size**2, size**2])
matte = tf.convert_to_tensor(value=matte_init)
laplacian = tf.convert_to_tensor(value=laplacian_init)
loss = matting.loss(matte, laplacian)
with self.subTest(name="matte"):
self.assert_jacobian_is_correct(matte, matte_init, loss)
with self.subTest(name="laplacian"):
self.assert_jacobian_is_correct(laplacian, laplacian_init, loss)
def test_loss_jacobian_random(self):
"""Tests the Jacobian of the matting loss function."""
shape = np.random.randint(3, 5, size=3)
matte_init = np.random.uniform(0.0, 1.0, size=shape.tolist() + [1])
shape[1:3] -= 2
weights_init = np.random.uniform(0.0,
1.0,
size=shape.tolist() + [9, 9])
matte = tf.convert_to_tensor(value=matte_init)
weights = tf.convert_to_tensor(value=weights_init)
loss = matting.loss(matte, weights)
with self.subTest(name="matte"):
self.assert_jacobian_is_correct(matte, matte_init, loss)
with self.subTest(name="weights"):
self.assert_jacobian_is_correct(weights, weights_init, loss)