class BeliefTests(unittest.TestCase):
def setUp(self):
self.model = LinearMRF(1, 2)
def test_belief_convergence(self):
new_beliefs = np.array([[0, 1], [0, 1]])
old_beliefs = np.array([[1, 0], [1, 0]])
result = self.model.check_convergence(new_beliefs, old_beliefs, 0.2)
self.assertEqual(result, False)
def test_pairwise_beliefs_shape(self):
beliefs = np.array([[0, 1], [1, 0]])
result = self.model.get_pairwise_beliefs(beliefs)
self.assertEqual(result.shape, (1, 4))
def test_pairwise_beliefs_value(self):
beliefs = np.array([[0, 1], [1, 0]])
result = self.model.get_pairwise_beliefs(beliefs)
np.testing.assert_array_equal(result, [[0, 0, 1, 0]])
def test_pairwise_beliefs_value_2x2(self):
self.model = LinearMRF(2, 2)
beliefs = np.array([[0, 1], [0, 1], [1, 0], [0, 1]])
result = self.model.get_pairwise_beliefs(beliefs)
np.testing.assert_array_equal(
result, [[0, 0, 0, 1], [0, 0, 1, 0], [0, 0, 0, 1], [0, 1, 0, 0]])
def main(_):
"""High level pipeline.
This script performs the trainsing, evaling and testing state of the model.
"""
learning_rate = FLAGS.learning_rate
num_epochs = FLAGS.num_epochs
convergence_margin = FLAGS.convergence_margin
input_file_path = FLAGS.input_file_path
# Load dataset.
original_img, noisy_samples = load_dataset(input_file_path, 10)
height = original_img.shape[0]
width = original_img.shape[1]
original_img = original_img.flatten()
noisy_samples = [sample.flatten() for sample in noisy_samples]
#print(original_img.shape)
#print(len(noisy_samples))
#exit()
# Initialize model.
model = LinearMRF(width, height)
model.train(original_img, noisy_samples, learning_rate, num_epochs,
convergence_margin)
# Evaluate model on training dataset
denoised_images = model.test(noisy_samples, convergence_margin)
# Plot inference result on image
plot_image(noisy_samples[0].reshape(height, width), 'Noisy Image',
'data/noisy_sample.png')
plot_image(denoised_images[0], 'Denoised Version', 'data/denoised_img.png')
def test_local_score(self):
self.model = LinearMRF(2, 2)
node = 1
assignment = 1
unary_beliefs = np.array([[0, 1], [0, 1], [1, 0], [0, 1]])
unary_potentials = np.array([[0.1, 0.8], [0.5, 0.71], [0.6, 0.2],
[1, 0]])
pairwise_potentials = np.array([[1, 1, 1, 0.2], [0, 0, 0, 0],
[1, 1, 0.3, 0.5], [1, 1, 1.2, 1]])
result = self.model.calculate_local_score(node, assignment,
unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(result, 0.71 + 0.2 + 0.5)
pairwise_potentials = np.array([[1, 1, 1, 0.7], [0, 0, 0, 0],
[1, 1, 1, 7], [1, 1, 1, 2]])
result = self.model.calculate_local_score(node, assignment,
unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(result, 0.71 + 0.7 + 7)
pairwise_potentials = np.array([[1, 1, 1, 5], [2, 2, 2, 2],
[1, 1, 1, 7], [1, 1, 1, -1]])
result = self.model.calculate_local_score(node, assignment,
unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(result, 0.71 + 5 + 7)
class ModelTests(unittest.TestCase):
def setUp(self):
self.model = LinearMRF(3, 2)
def test_unary_feature_shape(self):
img = np.array([1, 0, 0, 1, 0, 1])
result = self.model.get_unary_features(img)
self.assertEqual(result.shape, (6, 2))
def test_pairwise_feature_shape(self):
result = self.model.get_pairwise_features()
self.assertEqual(result.shape, (7, 4))
class LearningTests(unittest.TestCase):
def setUp(self):
self.model = LinearMRF(1, 2)
def test_learning_obj(self):
img_features = np.array([[1, 0], [1, 0]])
unary_beliefs = [tf.constant([[1, 0], [1, 0]])]
pair_beliefs = [tf.constant([[1, 0, 0, 0]])]
unary_potentials = [tf.constant([[1, 1], [1, 1]])]
pairwise_potentials = tf.constant([[1, 1, 1, 1]])
correct = 0
result = self.model.build_training_obj(img_features, unary_beliefs,
pair_beliefs, unary_potentials,
pairwise_potentials)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
result_val = sess.run(result)
self.assertEqual(correct, result_val)
def test_learning_obj2(self):
img_features = np.array([[1, 0], [1, 0], [0, 1]])
unary_beliefs = [
tf.constant([[1, 0], [0, 1], [0, 1]]),
tf.constant([[1, 0], [1, 0], [1, 0]])
]
pair_beliefs = [
tf.constant([[1, 0, 0, 0], [0, 1, 0, 0]]),
tf.constant([[1, 0, 0, 0], [0, 0, 0, 1]])
]
unary_potentials = [
tf.constant([[1, 1], [1, 1], [1, 1]]),
tf.constant([[1, 1], [1, 0], [1, 1]])
]
pairwise_potentials = tf.constant([[1, 1, 1, 1], [1, 1, 1, 1]])
correct = 4
result = self.model.build_training_obj(img_features, unary_beliefs,
pair_beliefs, unary_potentials,
pairwise_potentials)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
result_val = sess.run(result)
self.assertEqual(correct, result_val)
class InferenceTests(unittest.TestCase):
def setUp(self):
self.model = LinearMRF(1, 2)
def test_inf(self):
unary_beliefs = np.array([[0, 1], [0, 1]])
unary_potentials = np.array([[1, 0], [1, 0]])
pairwise_potentials = np.array([[2, 1, 1, 0]])
correct = np.array([[1, 0], [1, 0]])
result = self.model.inference_itr(unary_beliefs, unary_potentials,
pairwise_potentials)
np.testing.assert_array_equal(correct, result)
class ModelTests(unittest.TestCase):
def setUp(self):
self.model = LinearMRF(3, 2)
def test_unary_feature_shape(self):
img = np.array([1, 0, 0, 1, 0, 1])
result = self.model.get_unary_features(img)
self.assertEqual(result.shape, (6, 2))
def test_unary_feature_data(self):
img = np.array([1, 0, 0, 1, 0, 1])
result = self.model.get_unary_features(img)
expected = np.array([[0, 1], [1, 0], [1, 0], [0, 1], [1, 0], [0, 1]])
np.testing.assert_array_equal(result, expected)
def test_pairwise_feature_shape(self):
result = self.model.get_pairwise_features()
self.assertEqual(result.shape, (7, 4))
def test_pairwise_feature_data(self):
result = self.model.get_pairwise_features()
np.testing.assert_array_equal(result, [[1, 0, 0, 1]] * 7)
def setUp(self):
self.model = LinearMRF(3, 2)
class InferenceTests(unittest.TestCase):
def setUp(self):
self.model = LinearMRF(1, 2)
def test_inf(self):
unary_beliefs = np.array([[0, 1], [0, 1]])
unary_potentials = np.array([[1, 0], [1, 0]])
pairwise_potentials = np.array([[2, 1, 1, 0]])
correct = np.array([[1, 0], [1, 0]])
result = self.model.inference_itr(unary_beliefs, unary_potentials,
pairwise_potentials)
np.testing.assert_array_equal(correct, result)
def test_local_score_1x2_1(self):
unary_beliefs = np.array([[1, 0], [0, 1]])
unary_potentials = np.array([[1, 0], [1, 0]])
pairwise_potentials = np.array([[2, 1, 1, 0]])
result = self.model.calculate_local_score(1, 0, unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(3, result)
result = self.model.calculate_local_score(1, 1, unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(1, result)
def test_local_score_1x2_2(self):
unary_beliefs = np.array([[0, 1], [0, 1]])
unary_potentials = np.array([[1, 0], [1, 0]])
pairwise_potentials = np.array([[2, 1, 1, 0.5]])
result = self.model.calculate_local_score(0, 0, unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(1 + 1, result)
result = self.model.calculate_local_score(0, 1, unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(0.5, result)
def test_local_score(self):
self.model = LinearMRF(2, 2)
node = 1
assignment = 1
unary_beliefs = np.array([[0, 1], [0, 1], [1, 0], [0, 1]])
unary_potentials = np.array([[0.1, 0.8], [0.5, 0.71], [0.6, 0.2],
[1, 0]])
pairwise_potentials = np.array([[1, 1, 1, 0.2], [0, 0, 0, 0],
[1, 1, 0.3, 0.5], [1, 1, 1.2, 1]])
result = self.model.calculate_local_score(node, assignment,
unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(result, 0.71 + 0.2 + 0.5)
pairwise_potentials = np.array([[1, 1, 1, 0.7], [0, 0, 0, 0],
[1, 1, 1, 7], [1, 1, 1, 2]])
result = self.model.calculate_local_score(node, assignment,
unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(result, 0.71 + 0.7 + 7)
pairwise_potentials = np.array([[1, 1, 1, 5], [2, 2, 2, 2],
[1, 1, 1, 7], [1, 1, 1, -1]])
result = self.model.calculate_local_score(node, assignment,
unary_beliefs,
unary_potentials,
pairwise_potentials)
self.assertEqual(result, 0.71 + 5 + 7)
def test_pairwise_beliefs_value_2x2(self):
self.model = LinearMRF(2, 2)
beliefs = np.array([[0, 1], [0, 1], [1, 0], [0, 1]])
result = self.model.get_pairwise_beliefs(beliefs)
np.testing.assert_array_equal(
result, [[0, 0, 0, 1], [0, 0, 1, 0], [0, 0, 0, 1], [0, 1, 0, 0]])