def get_input_arrays(self):
"""Returns a list of the names of the input tensors.
Returns:
List of strings.
"""
return [_tensor_name(tensor) for tensor in self._input_tensors]
def get_input_arrays(self):
"""Returns a list of the names of the input tensors.
Returns:
List of strings.
"""
return [_tensor_name(tensor) for tensor in self._input_tensors]
def get_input_arrays(self):
"""Returns a list of the names of the input tensors.
Returns:
List of strings.
"""
if self._has_valid_tensors():
return [_tensor_name(tensor) for tensor in self._input_tensors]
else:
return [name for name, _ in self._input_arrays_with_shape]
def _freeze_graph(sess, output_tensors):
"""Returns a frozen GraphDef.
Freezes a graph with Variables in it. Otherwise the existing GraphDef is
returned.
Args:
sess: TensorFlow Session.
output_tensors: List of output tensors (only .name is used from this).
Returns:
Frozen GraphDef.
"""
if not _is_frozen_graph(sess):
output_arrays = [_tensor_name(tensor) for tensor in output_tensors]
return _tf_graph_util.convert_variables_to_constants(
sess, sess.graph_def, output_arrays)
else:
return sess.graph_def
def convert(self):
"""Converts a TensorFlow GraphDef based on instance variables.
Returns:
The converted data in serialized format. Either a TFLite Flatbuffer or a
Graphviz graph depending on value in `output_format`.
Raises:
ValueError:
Input shape is not specified.
None value for dimension in input_tensor.
"""
# Checks dimensions in input tensor.
if self._has_valid_tensors():
for tensor in self._input_tensors:
if not tensor.get_shape():
raise ValueError("Provide an input shape for input array "
"'{0}'.".format(_tensor_name(tensor)))
shape = tensor.get_shape().as_list()
if None in shape[1:]:
raise ValueError(
"None is only supported in the 1st dimension. Tensor '{0}' has "
"invalid shape '{1}'.".format(_tensor_name(tensor), shape))
elif shape[0] is None:
self._set_batch_size(batch_size=1)
# Get quantization stats. Ensures there is one stat per name if the stats
# are specified.
if self.quantized_input_stats:
quantized_stats = []
invalid_stats = []
for name in self.get_input_arrays():
if name in self.quantized_input_stats:
quantized_stats.append(self.quantized_input_stats[name])
else:
invalid_stats.append(name)
if invalid_stats:
raise ValueError("Quantization input stats are not available for input "
"tensors '{0}'.".format(",".join(invalid_stats)))
else:
quantized_stats = None
converter_kwargs = {
"inference_type": self.inference_type,
"inference_input_type": self.inference_input_type,
"input_format": constants.TENSORFLOW_GRAPHDEF,
"output_format": self.output_format,
"quantized_input_stats": quantized_stats,
"default_ranges_stats": self.default_ranges_stats,
"drop_control_dependency": self.drop_control_dependency,
"reorder_across_fake_quant": self.reorder_across_fake_quant,
"change_concat_input_ranges": self.change_concat_input_ranges,
"allow_custom_ops": self.allow_custom_ops,
"quantize_weights": self.quantize_weights,
"dump_graphviz_dir": self.dump_graphviz_dir,
"dump_graphviz_video": self.dump_graphviz_video
}
# Converts model.
if self._has_valid_tensors():
result = _toco_convert_impl(
input_data=self._graph_def,
input_tensors=self._input_tensors,
output_tensors=self._output_tensors,
**converter_kwargs)
else:
result = _toco_convert_graph_def(
input_data=self._graph_def,
input_arrays_with_shape=self._input_arrays_with_shape,
output_arrays=self._output_arrays,
**converter_kwargs)
return result
