def testEndpointsReuse(self):
inputs = create_test_input(2, 32, 32, 3)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points0 = xception.xception_65(inputs,
num_classes=10,
reuse=False)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points1 = xception.xception_65(inputs,
num_classes=10,
reuse=True)
self.assertItemsEqual(end_points0.keys(), end_points1.keys())
def testEndpointsReuse(self):
inputs = create_test_input(2, 32, 32, 3)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points0 = xception.xception_65(
inputs,
num_classes=10,
reuse=False)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points1 = xception.xception_65(
inputs,
num_classes=10,
reuse=True)
self.assertItemsEqual(end_points0.keys(), end_points1.keys())
def testAtrousFullyConvolutionalValues(self):
"""Verify dense feature extraction with atrous convolution."""
nominal_stride = 32
for output_stride in [4, 8, 16, 32, None]:
with slim.arg_scope(xception.xception_arg_scope()):
with tf.Graph().as_default():
with self.test_session() as sess:
tf.set_random_seed(0)
inputs = create_test_input(2, 96, 97, 3)
# Dense feature extraction followed by subsampling.
output, _ = self._xception_small(
inputs,
None,
is_training=False,
global_pool=False,
output_stride=output_stride)
if output_stride is None:
factor = 1
else:
factor = nominal_stride // output_stride
output = resnet_utils.subsample(output, factor)
# Make the two networks use the same weights.
tf.get_variable_scope().reuse_variables()
# Feature extraction at the nominal network rate.
expected, _ = self._xception_small(
inputs,
None,
is_training=False,
global_pool=False)
sess.run(tf.global_variables_initializer())
self.assertAllClose(output.eval(), expected.eval(),
atol=1e-5, rtol=1e-5)
def testFullyConvolutionalUnknownHeightWidth(self):
batch = 2
height, width = 65, 65
global_pool = False
inputs = create_test_input(batch, None, None, 3)
with slim.arg_scope(xception.xception_arg_scope()):
output, _ = self._xception_small(inputs,
None,
global_pool=global_pool)
self.assertListEqual(output.get_shape().as_list(),
[batch, None, None, 16])
images = create_test_input(batch, height, width, 3)
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
output = sess.run(output, {inputs: images.eval()})
self.assertEquals(output.shape, (batch, 3, 3, 16))
def testFullyConvolutionalUnknownHeightWidth(self):
batch = 2
height, width = 65, 65
global_pool = False
inputs = create_test_input(batch, None, None, 3)
with slim.arg_scope(xception.xception_arg_scope()):
output, _ = self._xception_small(
inputs,
None,
global_pool=global_pool)
self.assertListEqual(output.get_shape().as_list(),
[batch, None, None, 16])
images = create_test_input(batch, height, width, 3)
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
output = sess.run(output, {inputs: images.eval()})
self.assertEquals(output.shape, (batch, 3, 3, 16))
def testClassificationEndPoints(self):
global_pool = True
num_classes = 10
inputs = create_test_input(2, 224, 224, 3)
with slim.arg_scope(xception.xception_arg_scope()):
logits, end_points = self._xception_small(inputs,
num_classes=num_classes,
global_pool=global_pool,
scope='xception')
self.assertTrue(logits.op.name.startswith('xception/logits'))
self.assertListEqual(logits.get_shape().as_list(),
[2, 1, 1, num_classes])
self.assertTrue('predictions' in end_points)
self.assertListEqual(end_points['predictions'].get_shape().as_list(),
[2, 1, 1, num_classes])
self.assertTrue('global_pool' in end_points)
self.assertListEqual(end_points['global_pool'].get_shape().as_list(),
[2, 1, 1, 16])
def testUseBoundedAcitvation(self):
global_pool = False
num_classes = 10
output_stride = 8
for use_bounded_activation in (True, False):
tf.reset_default_graph()
inputs = create_test_input(2, 321, 321, 3)
with slim.arg_scope(xception.xception_arg_scope(
use_bounded_activation=use_bounded_activation)):
_, _ = self._xception_small(
inputs,
num_classes,
global_pool=global_pool,
output_stride=output_stride,
scope='xception')
for node in tf.get_default_graph().as_graph_def().node:
if node.op.startswith('Relu'):
self.assertEqual(node.op == 'Relu6', use_bounded_activation)
def testUseBoundedAcitvation(self):
global_pool = False
num_classes = 10
output_stride = 8
for use_bounded_activation in (True, False):
tf.reset_default_graph()
inputs = create_test_input(2, 321, 321, 3)
with slim.arg_scope(xception.xception_arg_scope(
use_bounded_activation=use_bounded_activation)):
_, _ = self._xception_small(
inputs,
num_classes,
global_pool=global_pool,
output_stride=output_stride,
scope='xception')
for node in tf.get_default_graph().as_graph_def().node:
if node.op.startswith('Relu'):
self.assertEqual(node.op == 'Relu6', use_bounded_activation)
def testClassificationEndPoints(self):
global_pool = True
num_classes = 10
inputs = create_test_input(2, 224, 224, 3)
with slim.arg_scope(xception.xception_arg_scope()):
logits, end_points = self._xception_small(
inputs,
num_classes=num_classes,
global_pool=global_pool,
scope='xception')
self.assertTrue(
logits.op.name.startswith('xception/logits'))
self.assertListEqual(logits.get_shape().as_list(), [2, 1, 1, num_classes])
self.assertTrue('predictions' in end_points)
self.assertListEqual(end_points['predictions'].get_shape().as_list(),
[2, 1, 1, num_classes])
self.assertTrue('global_pool' in end_points)
self.assertListEqual(end_points['global_pool'].get_shape().as_list(),
[2, 1, 1, 16])
def testUnknownBatchSize(self):
batch = 2
height, width = 65, 65
global_pool = True
num_classes = 10
inputs = create_test_input(None, height, width, 3)
with slim.arg_scope(xception.xception_arg_scope()):
logits, _ = self._xception_small(inputs,
num_classes,
global_pool=global_pool,
scope='xception')
self.assertTrue(logits.op.name.startswith('xception/logits'))
self.assertListEqual(logits.get_shape().as_list(),
[None, 1, 1, num_classes])
images = create_test_input(batch, height, width, 3)
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
output = sess.run(logits, {inputs: images.eval()})
self.assertEquals(output.shape, (batch, 1, 1, num_classes))
def testUnknownBatchSize(self):
batch = 2
height, width = 65, 65
global_pool = True
num_classes = 10
inputs = create_test_input(None, height, width, 3)
with slim.arg_scope(xception.xception_arg_scope()):
logits, _ = self._xception_small(
inputs,
num_classes,
global_pool=global_pool,
scope='xception')
self.assertTrue(logits.op.name.startswith('xception/logits'))
self.assertListEqual(logits.get_shape().as_list(),
[None, 1, 1, num_classes])
images = create_test_input(batch, height, width, 3)
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
output = sess.run(logits, {inputs: images.eval()})
self.assertEquals(output.shape, (batch, 1, 1, num_classes))
def testFullyConvolutionalEndpointShapes(self):
global_pool = False
num_classes = 10
inputs = create_test_input(2, 321, 321, 3)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points = self._xception_small(
inputs,
num_classes,
global_pool=global_pool,
scope='xception')
endpoint_to_shape = {
'xception/entry_flow/conv1_1': [2, 161, 161, 32],
'xception/entry_flow/block1': [2, 81, 81, 1],
'xception/entry_flow/block2': [2, 41, 41, 2],
'xception/entry_flow/block4': [2, 21, 21, 4],
'xception/middle_flow/block1': [2, 21, 21, 4],
'xception/exit_flow/block1': [2, 11, 11, 8],
'xception/exit_flow/block2': [2, 11, 11, 16]}
for endpoint, shape in six.iteritems(endpoint_to_shape):
self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)
def testClassificationShapes(self):
global_pool = True
num_classes = 10
inputs = create_test_input(2, 224, 224, 3)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points = self._xception_small(
inputs,
num_classes,
global_pool=global_pool,
scope='xception')
endpoint_to_shape = {
'xception/entry_flow/conv1_1': [2, 112, 112, 32],
'xception/entry_flow/block1': [2, 56, 56, 1],
'xception/entry_flow/block2': [2, 28, 28, 2],
'xception/entry_flow/block4': [2, 14, 14, 4],
'xception/middle_flow/block1': [2, 14, 14, 4],
'xception/exit_flow/block1': [2, 7, 7, 8],
'xception/exit_flow/block2': [2, 7, 7, 16]}
for endpoint, shape in six.iteritems(endpoint_to_shape):
self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)
def testAtrousFullyConvolutionalEndpointShapes(self):
global_pool = False
num_classes = 3
output_stride = 8
inputs = create_test_input(2, 65, 65, 3)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points = self._xception_small(
inputs,
num_classes,
global_pool=global_pool,
output_stride=output_stride,
scope='xception')
endpoint_to_shape = {
'xception/entry_flow/block1': [2, 17, 17, 1],
'xception/entry_flow/block2': [2, 9, 9, 2],
'xception/entry_flow/block4': [2, 9, 9, 4],
'xception/middle_flow/block1': [2, 9, 9, 4],
'xception/exit_flow/block1': [2, 9, 9, 8],
'xception/exit_flow/block2': [2, 9, 9, 16]}
for endpoint, shape in six.iteritems(endpoint_to_shape):
self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)
def testEndpointNames(self):
global_pool = True
num_classes = 10
inputs = create_test_input(2, 224, 224, 3)
with slim.arg_scope(xception.xception_arg_scope()):
_, end_points = self._xception_small(
inputs,
num_classes=num_classes,
global_pool=global_pool,
scope='xception')
expected = [
'xception/entry_flow/conv1_1',
'xception/entry_flow/conv1_2',
'xception/entry_flow/block1/unit_1/xception_module/separable_conv1',
'xception/entry_flow/block1/unit_1/xception_module/separable_conv2',
'xception/entry_flow/block1/unit_1/xception_module/separable_conv3',
'xception/entry_flow/block1/unit_1/xception_module/shortcut',
'xception/entry_flow/block1/unit_1/xception_module',
'xception/entry_flow/block1',
'xception/entry_flow/block2/unit_1/xception_module/separable_conv1',
'xception/entry_flow/block2/unit_1/xception_module/separable_conv2',
'xception/entry_flow/block2/unit_1/xception_module/separable_conv3',
'xception/entry_flow/block2/unit_1/xception_module/shortcut',
'xception/entry_flow/block2/unit_1/xception_module',
'xception/entry_flow/block2',
'xception/entry_flow/block3/unit_1/xception_module/separable_conv1',
'xception/entry_flow/block3/unit_1/xception_module/separable_conv2',
'xception/entry_flow/block3/unit_1/xception_module/separable_conv3',
'xception/entry_flow/block3/unit_1/xception_module/shortcut',
'xception/entry_flow/block3/unit_1/xception_module',
'xception/entry_flow/block3',
'xception/entry_flow/block4/unit_1/xception_module/separable_conv1',
'xception/entry_flow/block4/unit_1/xception_module/separable_conv2',
'xception/entry_flow/block4/unit_1/xception_module/separable_conv3',
'xception/entry_flow/block4/unit_1/xception_module/shortcut',
'xception/entry_flow/block4/unit_1/xception_module',
'xception/entry_flow/block4',
'xception/middle_flow/block1/unit_1/xception_module/separable_conv1',
'xception/middle_flow/block1/unit_1/xception_module/separable_conv2',
'xception/middle_flow/block1/unit_1/xception_module/separable_conv3',
'xception/middle_flow/block1/unit_1/xception_module',
'xception/middle_flow/block1/unit_2/xception_module/separable_conv1',
'xception/middle_flow/block1/unit_2/xception_module/separable_conv2',
'xception/middle_flow/block1/unit_2/xception_module/separable_conv3',
'xception/middle_flow/block1/unit_2/xception_module',
'xception/middle_flow/block1',
'xception/exit_flow/block1/unit_1/xception_module/separable_conv1',
'xception/exit_flow/block1/unit_1/xception_module/separable_conv2',
'xception/exit_flow/block1/unit_1/xception_module/separable_conv3',
'xception/exit_flow/block1/unit_1/xception_module/shortcut',
'xception/exit_flow/block1/unit_1/xception_module',
'xception/exit_flow/block1',
'xception/exit_flow/block2/unit_1/xception_module/separable_conv1',
'xception/exit_flow/block2/unit_1/xception_module/separable_conv2',
'xception/exit_flow/block2/unit_1/xception_module/separable_conv3',
'xception/exit_flow/block2/unit_1/xception_module',
'xception/exit_flow/block2',
'global_pool',
'xception/logits',
'predictions',
]
self.assertItemsEqual(end_points.keys(), expected)
sys.path.append(TF_API)
from deeplab.core import xception
from tensorflow.contrib.slim.nets import resnet_utils
slim = tf.contrib.slim
if __name__ == '__main__':
inputs = tf.random_normal([1, 224, 224, 3])
with slim.arg_scope(xception.xception_arg_scope()):
net, end_points = xception.xception_65(inputs,
num_classes=100,
is_training=False,
global_pool=True,
keep_prob=0.5,
output_stride=None,
regularize_depthwise=False,
multi_grid=[12,16,18],
reuse=None,
scope='xception_65')
writer = tf.summary.FileWriter("./logs", graph=tf.get_default_graph())