def test_convert_saved_model_skip_op_check(self):
self.create_unsupported_saved_model()
print(
glob.glob(
os.path.join(self._tmp_dir, SESSION_BUNDLE_MODEL_DIR, '*')))
tf_saved_model_conversion.convert_tf_saved_model(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR),
'Softmax',
os.path.join(self._tmp_dir, SAVED_MODEL_DIR),
skip_op_check=True)
weights = [{
'paths': ['group1-shard1of1.bin'],
'weights': [{
'shape': [2, 2, 2],
'name': 'Softmax',
'dtype': 'float32'
}]
}]
tfjs_path = os.path.join(self._tmp_dir, SAVED_MODEL_DIR)
# Check model.json and weights manifest.
with open(os.path.join(tfjs_path, 'model.json'), 'rt') as f:
model_json = json.load(f)
self.assertTrue(model_json['modelTopology'])
weights_manifest = model_json['weightsManifest']
self.assertEqual(weights_manifest, weights)
self.assertTrue(
glob.glob(os.path.join(self._tmp_dir, SAVED_MODEL_DIR,
'group*-*')))
def test_convert_saved_model_strip_debug_ops(self):
self.create_saved_model_with_debug_ops()
tf_saved_model_conversion.convert_tf_saved_model(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR),
'add',
os.path.join(self._tmp_dir, SAVED_MODEL_DIR),
strip_debug_ops=True)
weights = [{
'paths': ['group1-shard1of1.bin'],
'weights': [{
'dtype': 'float32',
'name': 'add',
'shape': [2, 2]
}]
}]
tfjs_path = os.path.join(self._tmp_dir, SAVED_MODEL_DIR)
# Check model.json and weights manifest.
with open(os.path.join(tfjs_path, 'model.json'), 'rt') as f:
model_json = json.load(f)
self.assertTrue(model_json['modelTopology'])
weights_manifest = model_json['weightsManifest']
self.assertEqual(weights_manifest, weights)
self.assertTrue(
glob.glob(os.path.join(self._tmp_dir, SAVED_MODEL_DIR,
'group*-*')))
def test_convert_saved_model(self):
self.create_saved_model()
print(
glob.glob(
os.path.join(self._tmp_dir, SESSION_BUNDLE_MODEL_DIR, '*')))
tf_saved_model_conversion.convert_tf_saved_model(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR), 'Softmax',
os.path.join(self._tmp_dir, SAVED_MODEL_DIR))
weights = [{
'paths': ['group1-shard1of1'],
'weights': [{
'shape': [2, 2],
'name': 'Softmax',
'dtype': 'float32'
}]
}]
# Load the saved weights as a JSON string.
weights_manifest = open(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR,
'weights_manifest.json'), 'rt')
output_json = json.load(weights_manifest)
weights_manifest.close()
self.assertEqual(output_json, weights)
# Check the content of the output directory.
self.assertTrue(
glob.glob(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR,
'tensorflowjs_model.pb')))
self.assertTrue(
glob.glob(os.path.join(self._tmp_dir, SAVED_MODEL_DIR,
'group*-*')))
def main():
parser = argparse.ArgumentParser('TensorFlow.js model converters.')
parser.add_argument(
'input_path',
type=str,
help='Path to the input file or directory. For input format "keras", '
'an HDF5 (.h5) file is expected. For input format "tensorflow", '
'a SavedModel or session bundle directory is expected.')
parser.add_argument(
'--input_format',
type=str,
required=True,
choices=set(['keras', 'tf_saved_model', 'tf_session_bundle']),
help='Input format. '
'For "keras", the input path can be one of the two following formats:\n'
' - A topology+weights combined HDF5 (e.g., generated with'
' `keras.model.save_model()` method).\n'
' - A weights-only HDF5 (e.g., generated with Keras Model\'s '
' `save_weights()` method). \n'
'For "tensorflow", a SavedModel or session bundle model is expected.')
parser.add_argument(
'--output_node_names',
type=str,
help='The names of the output nodes, separated by commas. E.g., '
'"logits,activations". Applicable only if input format is '
'"tf_saved_model" or "tf_session_bundle".')
parser.add_argument(
'--saved_model_tags',
type=str,
default='serve',
help='Tags of the MetaGraphDef to load, in comma separated string '
'format. Defaults to "serve". Applicable only if input format is '
'"tf_saved_model".')
parser.add_argument('output_dir',
type=str,
help='Path for all output artifacts.')
FLAGS = parser.parse_args()
# TODO(cais, piyu): More conversion logics can be added as additional
# branches below.
if FLAGS.input_format == 'keras':
if FLAGS.output_node_names:
raise ValueError(
'The --output_node_names flag is applicable only to input format '
'"tensorflow", but the current input format is "keras".')
dispatch_pykeras_conversion(FLAGS.input_path,
output_dir=FLAGS.output_dir)
elif FLAGS.input_format == 'tf_saved_model':
tf_saved_model_conversion.convert_tf_saved_model(
FLAGS.input_path,
FLAGS.output_node_names,
FLAGS.output_dir,
saved_model_tags=FLAGS.saved_model_tags)
elif FLAGS.input_format == 'tf_session_bundle':
tf_saved_model_conversion.convert_tf_session_bundle(
FLAGS.input_path, FLAGS.output_node_names, FLAGS.output_dir)
else:
raise ValueError('Invalid input format: \'%s\'' % FLAGS.input_format)
def test_optimizer_add_unsupported_op(self):
self.create_saved_model()
print(glob.glob(os.path.join(self._tmp_dir, SAVED_MODEL_DIR, '*')))
with self.assertRaisesRegexp( # pylint: disable=deprecated-method
ValueError, r'^Unsupported Ops'):
node = tf.test.mock.Mock(op='unknown')
graph = tf.test.mock.Mock(node=[node])
with tf.test.mock.patch.object(tf_optimizer,
'OptimizeGraph',
return_value=graph):
tf_saved_model_conversion.convert_tf_saved_model(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR), 'Softmax',
os.path.join(self._tmp_dir, SAVED_MODEL_DIR))
def test_convert_saved_model_strip_debug_ops(self):
self.create_saved_model_with_debug_ops()
tf_saved_model_conversion.convert_tf_saved_model(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR),
'add',
os.path.join(self._tmp_dir, SAVED_MODEL_DIR),
strip_debug_ops=True)
# Check the content of the output directory.
self.assertTrue(
glob.glob(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR,
'tensorflowjs_model.pb')))
self.assertTrue(
glob.glob(
os.path.join(self._tmp_dir, SAVED_MODEL_DIR, 'group*-*')))
def convert_to_tfjs(
savedmodel_dir,
output_dir,
tags,
signature,
inputs,
outputs,
quantization_dtype,
skip_op_check,
strip_debug_ops
):
with tf.Graph().as_default() as graph, tf.Session(graph=graph) as sess:
meta_graph = tf.saved_model.loader.load(sess, tags, savedmodel_dir)
meta = meta_graph.signature_def[signature]
output_node_names = [
meta.outputs[key].name for key in outputs
]
# Getting input/output tensor name from signature for information.
input_tensors = [
graph.get_tensor_by_name(meta.inputs[key].name) for key in inputs
]
output_tensors = [
graph.get_tensor_by_name(name) for name in output_node_names
]
print('input tensors:')
for i, t in enumerate(input_tensors):
print(' {}: "{}" {}'.format(i, t.name, t.shape.as_list()))
print('output tensors:')
for i, t in enumerate(output_tensors):
print(' {}: "{}" {}'.format(i, t.name, t.shape.as_list()))
tf_saved_model_conversion.convert_tf_saved_model(
savedmodel_dir,
output_node_names=','.join([n.split(':')[0] for n in output_node_names]),
output_dir=output_dir,
saved_model_tags=','.join(tags),
quantization_dtype=quantization_dtype,
skip_op_check=skip_op_check,
strip_debug_ops=strip_debug_ops)
def main():
parser = argparse.ArgumentParser('TensorFlow.js model converters.')
parser.add_argument(
'input_path',
nargs='?',
type=str,
help='Path to the input file or directory. For input format "keras", '
'an HDF5 (.h5) file is expected. For input format "tensorflow", '
'a SavedModel directory, session bundle directory, frozen model file, '
'or TF-Hub module is expected.')
parser.add_argument('output_path',
nargs='?',
type=str,
help='Path for all output artifacts.')
parser.add_argument(
'--input_format',
type=str,
required=False,
default='tf_saved_model',
choices=set([
'keras', 'keras_saved_model', 'tf_saved_model',
'tf_session_bundle', 'tf_frozen_model', 'tf_hub', 'tensorflowjs'
]),
help='Input format. '
'For "keras", the input path can be one of the two following formats:\n'
' - A topology+weights combined HDF5 (e.g., generated with'
' `keras.model.save_model()` method).\n'
' - A weights-only HDF5 (e.g., generated with Keras Model\'s '
' `save_weights()` method). \n'
'For "keras_saved_model", the input_path must point to a subfolder '
'under the saved model folder that is passed as the argument '
'to tf.contrib.save_model.save_keras_model(). '
'The subfolder is generated automatically by tensorflow when '
'saving tf.keras model in the SavedModel format. It is usually named '
'as a Unix epoch time (e.g., 1542212752).\n'
'For "tf" formats, a SavedModel, frozen model, session bundle model, '
' or TF-Hub module is expected.')
parser.add_argument('--output_format',
type=str,
required=False,
choices=set(['keras', 'tensorflowjs']),
default='tensorflowjs',
help='Output format. Default: tensorflowjs.')
parser.add_argument(
'--output_node_names',
type=str,
help='The names of the output nodes, separated by commas. E.g., '
'"logits,activations". Applicable only if input format is '
'"tf_saved_model" or "tf_session_bundle".')
parser.add_argument(
'--signature_name',
type=str,
help='Signature of the TF-Hub module to load. Applicable only if input'
' format is "tf_hub".')
parser.add_argument(
'--saved_model_tags',
type=str,
default='serve',
help='Tags of the MetaGraphDef to load, in comma separated string '
'format. Defaults to "serve". Applicable only if input format is '
'"tf_saved_model".')
parser.add_argument(
'--quantization_bytes',
type=int,
choices=set(quantization.QUANTIZATION_BYTES_TO_DTYPES.keys()),
help='How many bytes to optionally quantize/compress the weights to. 1- '
'and 2-byte quantizaton is supported. The default (unquantized) size is '
'4 bytes.')
parser.add_argument(
'--split_weights_by_layer',
action='store_true',
help='Applicable to keras input_format only: Whether the weights from '
'different layers are to be stored in separate weight groups, '
'corresponding to separate binary weight files. Default: False.')
parser.add_argument(
'--version',
'-v',
dest='show_version',
action='store_true',
help='Show versions of tensorflowjs and its dependencies')
parser.add_argument(
'--skip_op_check',
type=bool,
default=False,
help='Skip op validation for TensorFlow model conversion.')
parser.add_argument(
'--strip_debug_ops',
type=bool,
default=True,
help='Strip debug ops (Print, Assert, CheckNumerics) from graph.')
FLAGS = parser.parse_args()
if FLAGS.show_version:
print('\ntensorflowjs %s\n' % version.version)
print('Dependency versions:')
print(' keras %s' % keras.__version__)
print(' tensorflow %s' % tf.__version__)
return
if FLAGS.input_path is None:
raise ValueError('Error: The input_path argument must be set. '
'Run with --help flag for usage information.')
quantization_dtype = (
quantization.QUANTIZATION_BYTES_TO_DTYPES[FLAGS.quantization_bytes]
if FLAGS.quantization_bytes else None)
if (FLAGS.output_node_names and FLAGS.input_format
not in ('tf_saved_model', 'tf_session_bundle', 'tf_frozen_model')):
raise ValueError(
'The --output_node_names flag is applicable only to input formats '
'"tf_saved_model", "tf_session_bundle" and "tf_frozen_model", '
'but the current input format is "%s".' % FLAGS.input_format)
if FLAGS.signature_name and FLAGS.input_format != 'tf_hub':
raise ValueError(
'The --signature_name is applicable only to "tf_hub" input format, '
'but the current input format is "%s".' % FLAGS.input_format)
# TODO(cais, piyu): More conversion logics can be added as additional
# branches below.
if FLAGS.input_format == 'keras' and FLAGS.output_format == 'tensorflowjs':
dispatch_keras_h5_to_tensorflowjs_conversion(
FLAGS.input_path,
output_dir=FLAGS.output_path,
quantization_dtype=quantization_dtype,
split_weights_by_layer=FLAGS.split_weights_by_layer)
elif (FLAGS.input_format == 'keras_saved_model'
and FLAGS.output_format == 'tensorflowjs'):
dispatch_keras_saved_model_to_tensorflowjs_conversion(
FLAGS.input_path,
FLAGS.output_path,
quantization_dtype=quantization_dtype,
split_weights_by_layer=FLAGS.split_weights_by_layer)
elif (FLAGS.input_format == 'tf_saved_model'
and FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_saved_model(
FLAGS.input_path,
FLAGS.output_node_names,
FLAGS.output_path,
saved_model_tags=FLAGS.saved_model_tags,
quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tf_session_bundle'
and FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_session_bundle(
FLAGS.input_path,
FLAGS.output_node_names,
FLAGS.output_path,
quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tf_frozen_model'
and FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_frozen_model(
FLAGS.input_path,
FLAGS.output_node_names,
FLAGS.output_path,
quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tf_hub'
and FLAGS.output_format == 'tensorflowjs'):
if FLAGS.signature_name:
tf_saved_model_conversion.convert_tf_hub_module(
FLAGS.input_path,
FLAGS.output_path,
FLAGS.signature_name,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
else:
tf_saved_model_conversion.convert_tf_hub_module(
FLAGS.input_path,
FLAGS.output_path,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tensorflowjs'
and FLAGS.output_format == 'keras'):
dispatch_tensorflowjs_to_keras_h5_conversion(FLAGS.input_path,
FLAGS.output_path)
else:
raise ValueError(
'Unsupported input_format - output_format pair: %s - %s' %
(FLAGS.input_format, FLAGS.output_format))
def main():
parser = argparse.ArgumentParser('TensorFlow.js model converters.')
parser.add_argument(
'input_path',
nargs='?',
type=str,
help='Path to the input file or directory. For input format "keras", '
'an HDF5 (.h5) file is expected. For input format "tensorflow", '
'a SavedModel directory, session bundle directory, frozen model file, '
'or TF-Hub module is expected.')
parser.add_argument(
'output_path', nargs='?', type=str, help='Path for all output artifacts.')
parser.add_argument(
'--input_format',
type=str,
required=False,
default='tf_saved_model',
choices=set(['keras', 'tf_saved_model', 'tf_session_bundle',
'tf_frozen_model', 'tf_hub', 'tensorflowjs']),
help='Input format. '
'For "keras", the input path can be one of the two following formats:\n'
' - A topology+weights combined HDF5 (e.g., generated with'
' `keras.model.save_model()` method).\n'
' - A weights-only HDF5 (e.g., generated with Keras Model\'s '
' `save_weights()` method). \n'
'For "tf" formats, a SavedModel, frozen model, session bundle model, '
' or TF-Hub module is expected.')
parser.add_argument(
'--output_format',
type=str,
required=False,
choices=set(['keras', 'tensorflowjs']),
default='tensorflowjs',
help='Output format. Default: tensorflowjs.')
parser.add_argument(
'--output_node_names',
type=str,
help='The names of the output nodes, separated by commas. E.g., '
'"logits,activations". Applicable only if input format is '
'"tf_saved_model" or "tf_session_bundle".')
parser.add_argument(
'--signature_name',
type=str,
help='Signature of the TF-Hub module to load. Applicable only if input'
' format is "tf_hub".')
parser.add_argument(
'--saved_model_tags',
type=str,
default='serve',
help='Tags of the MetaGraphDef to load, in comma separated string '
'format. Defaults to "serve". Applicable only if input format is '
'"tf_saved_model".')
parser.add_argument(
'--quantization_bytes',
type=int,
choices=set(quantization.QUANTIZATION_BYTES_TO_DTYPES.keys()),
help='How many bytes to optionally quantize/compress the weights to. 1- '
'and 2-byte quantizaton is supported. The default (unquantized) size is '
'4 bytes.')
parser.add_argument(
'--version',
'-v',
dest='show_version',
action='store_true',
help='Show versions of tensorflowjs and its dependencies')
parser.add_argument(
'--skip_op_check',
type=bool,
default=False,
help='Skip op validation for TensorFlow model conversion.')
parser.add_argument(
'--strip_debug_ops',
type=bool,
default=True,
help='Strip debug ops (Print, Assert, CheckNumerics) from graph.')
FLAGS = parser.parse_args()
if FLAGS.show_version:
print('\ntensorflowjs %s\n' % version.version)
print('Dependency versions:')
print(' keras %s' % keras.__version__)
print(' tensorflow %s' % tf.__version__)
return
if FLAGS.input_path is None:
raise ValueError(
'Error: The input_path argument must be set. '
'Run with --help flag for usage information.')
quantization_dtype = (
quantization.QUANTIZATION_BYTES_TO_DTYPES[FLAGS.quantization_bytes]
if FLAGS.quantization_bytes else None)
if (FLAGS.output_node_names and
FLAGS.input_format not in
('tf_saved_model', 'tf_session_bundle', 'tf_frozen_model')):
raise ValueError(
'The --output_node_names flag is applicable only to input formats '
'"tf_saved_model", "tf_session_bundle" and "tf_frozen_model", '
'but the current input format is "%s".' % FLAGS.input_format)
if FLAGS.signature_name and FLAGS.input_format != 'tf_hub':
raise ValueError(
'The --signature_name is applicable only to "tf_hub" input format, '
'but the current input format is "%s".' % FLAGS.input_format)
# TODO(cais, piyu): More conversion logics can be added as additional
# branches below.
if FLAGS.input_format == 'keras' and FLAGS.output_format == 'tensorflowjs':
dispatch_keras_h5_to_tensorflowjs_conversion(
FLAGS.input_path, output_dir=FLAGS.output_path,
quantization_dtype=quantization_dtype)
elif (FLAGS.input_format == 'tf_saved_model' and
FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_saved_model(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, saved_model_tags=FLAGS.saved_model_tags,
quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tf_session_bundle' and
FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_session_bundle(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tf_frozen_model' and
FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_frozen_model(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tf_hub' and
FLAGS.output_format == 'tensorflowjs'):
if FLAGS.signature_name:
tf_saved_model_conversion.convert_tf_hub_module(
FLAGS.input_path, FLAGS.output_path, FLAGS.signature_name,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
else:
tf_saved_model_conversion.convert_tf_hub_module(
FLAGS.input_path,
FLAGS.output_path,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (FLAGS.input_format == 'tensorflowjs' and
FLAGS.output_format == 'keras'):
dispatch_tensorflowjs_to_keras_h5_conversion(FLAGS.input_path,
FLAGS.output_path)
else:
raise ValueError(
'Unsupported input_format - output_format pair: %s - %s' %
(FLAGS.input_format, FLAGS.output_format))
def main():
parser = argparse.ArgumentParser('TensorFlow.js model converters.')
parser.add_argument(
'input_path',
nargs='?',
type=str,
help='Path to the input file or directory. For input format "keras", '
'an HDF5 (.h5) file is expected. For input format "tensorflow", '
'a SavedModel directory, session bundle directory, frozen model file, '
'or TF-Hub module is expected.')
parser.add_argument(
'output_path', nargs='?', type=str, help='Path for all output artifacts.')
parser.add_argument(
'--input_format',
type=str,
required=False,
default='tf_saved_model',
choices=set(['keras', 'tf_saved_model', 'tf_session_bundle',
'tf_frozen_model', 'tf_hub', 'tensorflowjs']),
help='Input format. '
'For "keras", the input path can be one of the two following formats:\n'
' - A topology+weights combined HDF5 (e.g., generated with'
' `keras.model.save_model()` method).\n'
' - A weights-only HDF5 (e.g., generated with Keras Model\'s '
' `save_weights()` method). \n'
'For "tf" formats, a SavedModel, frozen model, session bundle model, '
' or TF-Hub module is expected.')
parser.add_argument(
'--output_format',
type=str,
required=False,
choices=set(['keras', 'tensorflowjs']),
default='tensorflowjs',
help='Output format. Default: tensorflowjs.')
parser.add_argument(
'--output_node_names',
type=str,
help='The names of the output nodes, separated by commas. E.g., '
'"logits,activations". Applicable only if input format is '
'"tf_saved_model" or "tf_session_bundle".')
parser.add_argument(
'--signature_name',
type=str,
help='Signature of the TF-Hub module to load. Applicable only if input'
' format is "tf_hub".')
parser.add_argument(
'--saved_model_tags',
type=str,
default='serve',
help='Tags of the MetaGraphDef to load, in comma separated string '
'format. Defaults to "serve". Applicable only if input format is '
'"tf_saved_model".')
parser.add_argument(
'--quantization_bytes',
type=int,
choices=set(quantization.QUANTIZATION_BYTES_TO_DTYPES.keys()),
help='How many bytes to optionally quantize/compress the weights to. 1- '
'and 2-byte quantizaton is supported. The default (unquantized) size is '
'4 bytes.')
parser.add_argument(
'--version',
'-v',
dest='show_version',
action='store_true',
help='Show versions of tensorflowjs and its dependencies')
FLAGS = parser.parse_args()
if FLAGS.show_version:
print('\ntensorflowjs %s\n' % version.version)
print('Dependency versions:')
print(' keras %s' % keras.__version__)
print(' tensorflow %s' % tf.__version__)
return
quantization_dtype = (
quantization.QUANTIZATION_BYTES_TO_DTYPES[FLAGS.quantization_bytes]
if FLAGS.quantization_bytes else None)
if (FLAGS.output_node_names and
FLAGS.input_format not in
('tf_saved_model', 'tf_session_bundle', 'tf_frozen_model')):
raise ValueError(
'The --output_node_names flag is applicable only to input formats '
'"tf_saved_model", "tf_session_bundle" and "tf_frozen_model", '
'but the current input format is "%s".' % FLAGS.input_format)
if FLAGS.signature_name and FLAGS.input_format != 'tf_hub':
raise ValueError(
'The --signature_name is applicable only to "tf_hub" input format, '
'but the current input format is "%s".' % FLAGS.input_format)
# TODO(cais, piyu): More conversion logics can be added as additional
# branches below.
if FLAGS.input_format == 'keras' and FLAGS.output_format == 'tensorflowjs':
dispatch_keras_h5_to_tensorflowjs_conversion(
FLAGS.input_path, output_dir=FLAGS.output_path,
quantization_dtype=quantization_dtype)
elif (FLAGS.input_format == 'tf_saved_model' and
FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_saved_model(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, saved_model_tags=FLAGS.saved_model_tags,
quantization_dtype=quantization_dtype)
elif (FLAGS.input_format == 'tf_session_bundle' and
FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_session_bundle(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, quantization_dtype=quantization_dtype)
elif (FLAGS.input_format == 'tf_frozen_model' and
FLAGS.output_format == 'tensorflowjs'):
tf_saved_model_conversion.convert_tf_frozen_model(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, quantization_dtype=quantization_dtype)
elif (FLAGS.input_format == 'tf_hub' and
FLAGS.output_format == 'tensorflowjs'):
if FLAGS.signature_name:
tf_saved_model_conversion.convert_tf_hub_module(
FLAGS.input_path, FLAGS.output_path, FLAGS.signature_name)
else:
tf_saved_model_conversion.convert_tf_hub_module(FLAGS.input_path,
FLAGS.output_path)
elif (FLAGS.input_format == 'tensorflowjs' and
FLAGS.output_format == 'keras'):
dispatch_tensorflowjs_to_keras_h5_conversion(FLAGS.input_path,
FLAGS.output_path)
else:
raise ValueError(
'Unsupported input_format - output_format pair: %s - %s' %
(FLAGS.input_format, FLAGS.output_format))
def main():
FLAGS = setup_arugments()
if FLAGS.show_version:
print('\ntensorflowjs %s\n' % version.version)
print('Dependency versions:')
print(' keras %s' % keras.__version__)
print(' tensorflow %s' % tf.__version__)
return
if FLAGS.input_path is None:
raise ValueError(
'Error: The input_path argument must be set. '
'Run with --help flag for usage information.')
input_format, output_format = _standardize_input_output_formats(
FLAGS.input_format, FLAGS.output_format)
quantization_dtype = (
quantization.QUANTIZATION_BYTES_TO_DTYPES[FLAGS.quantization_bytes]
if FLAGS.quantization_bytes else None)
if (FLAGS.output_node_names and
input_format not in
('tf_saved_model', 'tf_session_bundle', 'tf_frozen_model')):
raise ValueError(
'The --output_node_names flag is applicable only to input formats '
'"tf_saved_model", "tf_session_bundle" and "tf_frozen_model", '
'but the current input format is "%s".' % FLAGS.input_format)
if FLAGS.signature_name and input_format != 'tf_hub':
raise ValueError(
'The --signature_name is applicable only to "tf_hub" input format, '
'but the current input format is "%s".' % input_format)
# TODO(cais, piyu): More conversion logics can be added as additional
# branches below.
if input_format == 'keras' and output_format == 'tfjs_layers_model':
dispatch_keras_h5_to_tensorflowjs_conversion(
FLAGS.input_path, output_dir=FLAGS.output_path,
quantization_dtype=quantization_dtype,
split_weights_by_layer=FLAGS.split_weights_by_layer)
elif (input_format == 'keras_saved_model' and
output_format == 'tfjs_layers_model'):
dispatch_keras_saved_model_to_tensorflowjs_conversion(
FLAGS.input_path, FLAGS.output_path,
quantization_dtype=quantization_dtype,
split_weights_by_layer=FLAGS.split_weights_by_layer)
elif (input_format == 'tf_saved_model' and
output_format == 'tfjs_graph_model'):
tf_saved_model_conversion.convert_tf_saved_model(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, saved_model_tags=FLAGS.saved_model_tags,
quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (input_format == 'tf_session_bundle' and
output_format == 'tfjs_graph_model'):
tf_saved_model_conversion.convert_tf_session_bundle(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (input_format == 'tf_frozen_model' and
output_format == 'tfjs_graph_model'):
tf_saved_model_conversion.convert_tf_frozen_model(
FLAGS.input_path, FLAGS.output_node_names,
FLAGS.output_path, quantization_dtype=quantization_dtype,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (input_format == 'tf_hub' and
output_format == 'tfjs_graph_model'):
if FLAGS.signature_name:
tf_saved_model_conversion.convert_tf_hub_module(
FLAGS.input_path, FLAGS.output_path, FLAGS.signature_name,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
else:
tf_saved_model_conversion.convert_tf_hub_module(
FLAGS.input_path,
FLAGS.output_path,
skip_op_check=FLAGS.skip_op_check,
strip_debug_ops=FLAGS.strip_debug_ops)
elif (input_format == 'tfjs_layers_model' and
output_format == 'keras'):
dispatch_tensorflowjs_to_keras_h5_conversion(FLAGS.input_path,
FLAGS.output_path)
else:
raise ValueError(
'Unsupported input_format - output_format pair: %s - %s' %
(input_format, output_format))
