def horizontal_lstm(images, num_filters_out, scope=None):
"""Run an LSTM bidirectionally over all the rows of each image.
Args:
images: (num_images, height, width, depth) tensor
num_filters_out: output depth
scope: optional scope name
Returns:
(num_images, height, width, num_filters_out) tensor, where
num_steps is width and new num_batches is num_image_batches * height
"""
with variable_scope.variable_scope(scope, "HorizontalLstm", [images]):
batch_size, _, _, _ = _shape(images)
sequence = images_to_sequence(images)
with variable_scope.variable_scope("lr"):
hidden_sequence_lr = lstm1d.ndlstm_base(sequence,
num_filters_out // 2)
with variable_scope.variable_scope("rl"):
hidden_sequence_rl = (lstm1d.ndlstm_base(sequence,
num_filters_out -
num_filters_out // 2,
reverse=1))
output_sequence = array_ops.concat(
[hidden_sequence_lr, hidden_sequence_rl], 2)
output = sequence_to_images(output_sequence, batch_size)
return output
def horizontal_lstm(images, num_filters_out, scope=None):
"""Run an LSTM bidirectionally over all the rows of each image.
Args:
images: (num_images, height, width, depth) tensor
num_filters_out: output depth
scope: optional scope name
Returns:
(num_images, height, width, num_filters_out) tensor, where
num_steps is width and new num_batches is num_image_batches * height
"""
with tf.variable_scope(scope, "HorizontalLstm", [images]):
batch_size, _, _, _ = _shape(images)
sequence = images_to_sequence(images)
with tf.variable_scope("lr"):
hidden_sequence_lr = lstm1d.ndlstm_base(sequence, num_filters_out // 2)
with tf.variable_scope("rl"):
hidden_sequence_rl = (
lstm1d.ndlstm_base(sequence,
num_filters_out - num_filters_out // 2,
reverse=1))
output_sequence = tf.concat_v2([hidden_sequence_lr, hidden_sequence_rl], 2)
output = sequence_to_images(output_sequence, batch_size)
return output
def testSequenceToSequenceDims(self):
with self.test_session():
inputs = tf.constant(_rand(17, 1, 5))
outputs = lstm1d.ndlstm_base(inputs, 8)
tf.initialize_all_variables().run()
result = outputs.eval()
self.assertEqual(tuple(result.shape), (17, 1, 8))
def testSequenceToSequenceGradientReverse(self):
with self.test_session():
size = (17, 1, 15)
output_size = (17, 1, 8)
inputs = tf.constant(_rand(*size))
outputs = lstm1d.ndlstm_base(inputs, 8, reverse=1, dynamic=False)
tf.initialize_all_variables().run()
if 1: # pylint: disable=using-constant-test
gradients = tf.gradients(outputs, inputs)[0].eval()
self.assertEqual(gradients.shape, size)
else:
# TODO(tmb) tf.test.compute_gradient error is currently broken
# with dynamic_rnn. Enable this test case eventually.
err = tf.test.compute_gradient_error(inputs,
size,
outputs,
output_size,
delta=1e-4)
self.assert_(not np.isnan(err))
self.assert_(err < 0.1)
def testSequenceToSequenceGradientReverse(self):
with self.test_session():
size = (17, 1, 15)
output_size = (17, 1, 8)
inputs = tf.constant(_rand(*size))
outputs = lstm1d.ndlstm_base(inputs, 8, reverse=1, dynamic=False)
tf.initialize_all_variables().run()
if 1: # pylint: disable=using-constant-test
gradients = tf.gradients(outputs, inputs)[0].eval()
self.assertEqual(gradients.shape, size)
else:
# TODO(tmb) tf.test.compute_gradient error is currently broken
# with dynamic_rnn. Enable this test case eventually.
err = tf.test.compute_gradient_error(inputs,
size,
outputs,
output_size,
delta=1e-4)
self.assert_(not np.isnan(err))
self.assert_(err < 0.1)