def testCalibrationLayerRegularization(self):
with ops.Graph().as_default():
# Shape: [batch_size=2, 3, 2]
uncalibrated = array_ops.constant([
[[0.1, 0.2], [0.9, 0.8], [0.4, 0.6]],
[[0.2, 0.3], [1.0, 0.9], [0.5, 0.7]],
])
kp_init_0 = self._UniformKeypoints(10)
kp_init_1 = self._UniformKeypoints(5, 0, 1000)
num_keypoints = [10, 10, 10, 10, 5, 5]
kp_init = [
kp_init_0, kp_init_0, kp_init_0, kp_init_0, kp_init_1,
kp_init_1
]
_, _, regularization = (pwl_calibration_layers.calibration_layer(
uncalibrated,
num_keypoints,
keypoints_initializers=kp_init,
l1_reg=0.1,
l2_reg=1.0,
l1_laplacian_reg=[0.3, 0.1, 0.2, 0.3, 0.4, 0.5],
l2_laplacian_reg=[None, 1.0, None, None, None, None],
name='test'))
with self.test_session() as sess:
sess.run(variables.global_variables_initializer())
got = sess.run(regularization)
expected_value = 6294341.5
self.assertAlmostEqual(got, expected_value, delta=1e-1)
def testCalibrationLayerWithUnknownBatchSize(self):
with ops.Graph().as_default():
# Shape: [batch_size=2, 3, 2]
uncalibrated = array_ops.placeholder(dtypes.float32,
shape=[None, 3, 2])
kp_init_0 = self._UniformKeypoints(10)
kp_init_1 = self._UniformKeypoints(5, 0, 1000)
num_keypoints = [10, 10, 10, 10, 5, 5]
kp_init = [
kp_init_0, kp_init_0, kp_init_0, kp_init_0, kp_init_1,
kp_init_1
]
calibrated, projection_ops, regularization = (
pwl_calibration_layers.calibration_layer(
uncalibrated,
num_keypoints,
keypoints_initializers=kp_init,
name='test'))
self.assertEqual(projection_ops, [])
self.assertEqual(regularization, None)
got = keypoints_initialization._materialize_locally(
calibrated,
num_steps=1,
feed_dict={
uncalibrated: [
[[0.1, 0.2], [0.9, 0.8], [0.4, 0.6]],
[[0.2, 0.3], [1.0, 0.9], [0.5, 0.7]],
]
})
want = [
[[210., 220.], [290., 280.], [400., 600.]],
[[220., 230.], [300., 290.], [500., 700.]],
]
self.assertAllClose(got, want)
def testCalibrationLayer(self):
with tf.Graph().as_default():
# Shape: [batch_size=2, 3, 2]
uncalibrated = tf.constant([
[[0.1, 0.2], [0.9, 0.8], [0.4, 0.6]],
[[0.2, 0.3], [1.0, 0.9], [0.5, 0.7]],
])
kp_init_0 = self._UniformKeypoints(10)
kp_init_1 = self._UniformKeypoints(5, 0, 1000)
num_keypoints = [10, 10, 10, 10, 5, 5]
kp_init = [
kp_init_0, kp_init_0, kp_init_0, kp_init_0, kp_init_1,
kp_init_1
]
calibrated, projection_ops, regularization = (
pwl_calibration_layers.calibration_layer(
uncalibrated,
num_keypoints,
keypoints_initializers=kp_init,
name='test'))
self.assertEqual(projection_ops, [])
self.assertEqual(regularization, None)
got = keypoints_initialization._materialize_locally(calibrated,
num_steps=1)
want = [
[[210., 220.], [290., 280.], [400., 600.]],
[[220., 230.], [300., 290.], [500., 700.]],
]
self.assertAllClose(got, want)
def _calibration_layer(input_tensor, input_dim, input_min, input_max,
num_keypoints, output_min, output_max):
"""Create an intermediate calibration layer."""
init_keypoints = keypoints_initialization.uniform_keypoints_for_signal(
num_keypoints=num_keypoints,
input_min=input_min,
input_max=input_max,
output_min=output_min,
output_max=output_max,
dtype=input_tensor.dtype)
packed_results = pwl_calibration_layers.calibration_layer(
input_tensor,
num_keypoints=num_keypoints,
keypoints_initializers=[init_keypoints] * input_dim,
bound=True,
monotonic=+1)
(calibrated_input_tensor, projection_ops, _) = packed_results
return (calibrated_input_tensor, projection_ops)