def generate_back_converters(converters):
back_converters = {}
for target_name, converter in six.iteritems(converters):
if isinstance(converter, partial
) and converter.func == tf_pb_to_tf_py.generic_converter:
from_name = converter.keywords['target_name']
revert_inputs = converter.keywords.get('revert_inputs', False)
attrib_name_dict = utils.key_value_swapped(
converter.keywords.get('attrib_name_dict', {}))
attrib_to_input_dict = utils.key_value_swapped(
converter.keywords.get('input_to_attrib_dict', {}))
attribs_to_remove = list(
six.iterkeys(converter.keywords.get('new_attribs', {})))
back_converters[from_name] = partial(
generic_converter,
target_name=target_name,
revert_inputs=revert_inputs,
attrib_name_dict=attrib_name_dict,
attrib_to_input_dict=attrib_to_input_dict,
attribs_to_remove=attribs_to_remove)
return back_converters
from __future__ import division, print_function, absolute_import
import typing
from functools import partial
import numpy as np
import six
from nnef_tools.conversion.tensorflow import tf_pb_to_tf_py
from nnef_tools.core import utils
from nnef_tools.io.tensorflow.tf_graph import *
from nnef_tools.shape_inference import shape_inference as infer
# noinspection PyProtectedMember
_tf_py_dtype_to_tf_pb_dtype = utils.key_value_swapped(
tf_pb_to_tf_py._tf_py_dtype_by_tf_pb_dtype)
def convert(tf_graph):
# type: (TFGraph)->None
for tensor in tf_graph.tensors:
if tensor.is_variable:
if tensor.data.dtype == np.int64:
tensor.data = tensor.data.astype(np.int32)
tensor.dtype = "int32"
if tensor.data.dtype == np.float64:
tensor.data = tensor.data.astype(np.float32)
tensor.dtype = "float32"