def arguments_post_parsing(argv: argparse.Namespace):
moc_front_end, available_moc_front_ends = get_moc_frontends(argv)
is_tf, is_caffe, is_mxnet, is_kaldi, is_onnx =\
deduce_framework_by_namespace(argv) if not moc_front_end else [False, False, False, False, False]
if not any([is_tf, is_caffe, is_mxnet, is_kaldi, is_onnx]):
frameworks = ['tf', 'caffe', 'mxnet', 'kaldi', 'onnx']
frameworks = list(set(frameworks + available_moc_front_ends))
if argv.framework not in frameworks:
if argv.use_legacy_frontend:
raise Error(
'Framework {} is not a valid target when using the --use_legacy_frontend flag. '
'The following legacy frameworks are available: {}' +
refer_to_faq_msg(15), argv.framework, frameworks)
else:
raise Error(
'Framework {} is not a valid target. Please use --framework with one from the list: {}. '
+ refer_to_faq_msg(15), argv.framework, frameworks)
if is_tf and not argv.input_model and not argv.saved_model_dir and not argv.input_meta_graph:
raise Error(
'Path to input model or saved model dir is required: use --input_model, --saved_model_dir or '
'--input_meta_graph')
elif is_mxnet and not argv.input_model and not argv.input_symbol and not argv.pretrained_model_name:
raise Error(
'Path to input model or input symbol or pretrained_model_name is required: use --input_model or '
'--input_symbol or --pretrained_model_name')
elif is_caffe and not argv.input_model and not argv.input_proto:
raise Error(
'Path to input model or input proto is required: use --input_model or --input_proto'
)
elif (is_kaldi or is_onnx) and not argv.input_model:
raise Error('Path to input model is required: use --input_model.')
log.debug(str(argv))
log.debug("Model Optimizer started")
model_name = "<UNKNOWN_NAME>"
if argv.model_name:
model_name = argv.model_name
elif argv.input_model:
model_name = get_model_name(argv.input_model)
elif is_tf and argv.saved_model_dir:
model_name = "saved_model"
elif is_tf and argv.input_meta_graph:
model_name = get_model_name(argv.input_meta_graph)
elif is_mxnet and argv.input_symbol:
model_name = get_model_name(argv.input_symbol)
argv.model_name = model_name
log.debug('Output model name would be {}{{.xml, .bin}}'.format(
argv.model_name))
# if --input_proto is not provided, try to retrieve another one
# by suffix substitution from model file name
if is_caffe and not argv.input_proto:
argv.input_proto = replace_ext(argv.input_model, '.caffemodel',
'.prototxt')
if not argv.input_proto:
raise Error(
"Cannot find prototxt file: for Caffe please specify --input_proto - a "
+
"protobuf file that stores topology and --input_model that stores "
+ "pretrained weights. " + refer_to_faq_msg(20))
log.info('Deduced name for prototxt: {}'.format(argv.input_proto))
if not argv.silent:
print_argv(argv, is_caffe, is_tf, is_mxnet, is_kaldi, is_onnx,
argv.model_name)
# This try-except is additional reinsurance that the IE
# dependency search does not break the MO pipeline
def raise_ie_not_found():
raise Error(
"Could not find the Inference Engine or nGraph Python API.\n"
"Consider building the Inference Engine and nGraph Python APIs from sources or try to install OpenVINO (TM) Toolkit using \"install_prerequisites.{}\""
.format("bat" if sys.platform == "windows" else "sh"))
try:
if not find_ie_version(silent=argv.silent):
raise_ie_not_found()
except Exception as e:
raise_ie_not_found()
if 'data_type' in argv and argv.data_type in ['FP16', 'half']:
argv.data_type = 'FP32'
argv.compress_fp16 = True
else:
argv.compress_fp16 = False
# This is just to check that transform key is valid and transformations are available
check_available_transforms(parse_transform(argv.transform))
if argv.legacy_ir_generation and len(argv.transform) != 0:
raise Error(
"--legacy_ir_generation and --transform keys can not be used at the same time."
)
# For C++ frontends there are no specific Python installation requirements, check only generic ones
if moc_front_end:
ret_code = check_requirements()
else:
ret_code = check_requirements(framework=argv.framework)
if ret_code:
raise Error(
'check_requirements exited with return code {}'.format(ret_code))
if is_tf and argv.tensorflow_use_custom_operations_config is not None:
argv.transformations_config = argv.tensorflow_use_custom_operations_config
if is_caffe and argv.mean_file and argv.mean_values:
raise Error(
'Both --mean_file and mean_values are specified. Specify either mean file or mean values. '
+ refer_to_faq_msg(17))
elif is_caffe and argv.mean_file and argv.mean_file_offsets:
values = get_tuple_values(argv.mean_file_offsets,
t=int,
num_exp_values=2)
mean_file_offsets = np.array([int(x) for x in values[0].split(',')])
if not all([offset >= 0 for offset in mean_file_offsets]):
raise Error(
"Negative value specified for --mean_file_offsets option. "
"Please specify positive integer values in format '(x,y)'. " +
refer_to_faq_msg(18))
argv.mean_file_offsets = mean_file_offsets
if argv.scale and argv.scale_values:
raise Error(
'Both --scale and --scale_values are defined. Specify either scale factor or scale values per input '
+ 'channels. ' + refer_to_faq_msg(19))
if argv.scale and argv.scale < 1.0:
log.error(
"The scale value is less than 1.0. This is most probably an issue because the scale value specifies "
"floating point value which all input values will be *divided*.",
extra={'is_warning': True})
if argv.input_model and (is_tf and argv.saved_model_dir):
raise Error('Both --input_model and --saved_model_dir are defined. '
'Specify either input model or saved model directory.')
if is_tf:
if argv.saved_model_tags is not None:
if ' ' in argv.saved_model_tags:
raise Error(
'Incorrect saved model tag was provided. Specify --saved_model_tags with no spaces in it'
)
argv.saved_model_tags = argv.saved_model_tags.split(',')
argv.output = argv.output.split(',') if argv.output else None
argv.placeholder_shapes, argv.placeholder_data_types = get_placeholder_shapes(
argv.input, argv.input_shape, argv.batch)
mean_values = parse_tuple_pairs(argv.mean_values)
scale_values = parse_tuple_pairs(argv.scale_values)
mean_scale = get_mean_scale_dictionary(mean_values, scale_values,
argv.input)
argv.mean_scale_values = mean_scale
if not os.path.exists(argv.output_dir):
try:
os.makedirs(argv.output_dir)
except PermissionError as e:
raise Error(
"Failed to create directory {}. Permission denied! " +
refer_to_faq_msg(22), argv.output_dir) from e
else:
if not os.access(argv.output_dir, os.W_OK):
raise Error(
"Output directory {} is not writable for current user. " +
refer_to_faq_msg(22), argv.output_dir)
log.debug("Placeholder shapes : {}".format(argv.placeholder_shapes))
if hasattr(argv,
'extensions') and argv.extensions and argv.extensions != '':
extensions = argv.extensions.split(',')
else:
extensions = None
argv.freeze_placeholder_with_value, argv.input = get_freeze_placeholder_values(
argv.input, argv.freeze_placeholder_with_value)
if is_tf:
from mo.front.tf.register_custom_ops import get_front_classes
import_extensions.load_dirs(argv.framework, extensions,
get_front_classes)
elif is_caffe:
send_framework_info('caffe')
from mo.front.caffe.register_custom_ops import get_front_classes
import_extensions.load_dirs(argv.framework, extensions,
get_front_classes)
elif is_mxnet:
send_framework_info('mxnet')
from mo.front.mxnet.register_custom_ops import get_front_classes
import_extensions.load_dirs(argv.framework, extensions,
get_front_classes)
elif is_kaldi:
send_framework_info('kaldi')
from mo.front.kaldi.register_custom_ops import get_front_classes
import_extensions.load_dirs(argv.framework, extensions,
get_front_classes)
elif is_onnx:
send_framework_info('onnx')
from mo.front.onnx.register_custom_ops import get_front_classes
import_extensions.load_dirs(argv.framework, extensions,
get_front_classes)
return argv
def load(self, graph: Graph):
argv = graph.graph['cmd_params']
if argv.tensorflow_custom_layer_libraries:
libraries = argv.tensorflow_custom_layer_libraries.split(',')
for library in libraries:
log.info('Loading library "{}" with custom operations'.format(
library))
tf_v1.load_op_library(library)
graph_def, variables_values, framework = load_tf_graph_def(
graph_file_name=argv.input_model,
is_binary=not argv.input_model_is_text,
checkpoint=argv.input_checkpoint,
user_output_node_names_list=argv.output,
model_dir=argv.saved_model_dir,
meta_graph_file=argv.input_meta_graph,
saved_model_tags=argv.saved_model_tags)
send_framework_info(framework)
try:
tf_v1.import_graph_def(graph_def, name='')
except:
log.warning(
"TensorFlow post-processing of loaded model was unsuccessful. "
"This is an optional step that Model Optimizer performs for any input model but it is not usually "
"required for all models. "
"It likely means that the original model is ill-formed. "
"Model Optimizer will continue converting this model.")
log.debug("Number of nodes in graph_def: {}".format(len(
graph_def.node))) # pylint: disable=no-member
if argv.tensorboard_logdir:
tensorboard_util.dump_for_tensorboard(graph_def,
argv.tensorboard_logdir)
update_extractors_with_extensions(tf_op_extractors)
try:
protobuf2nx(graph, graph_def)
except Exception as e:
raise Error(
'Cannot pre-process TensorFlow graph after reading from model file "{}". ' \
'File is corrupt or has unsupported format. Details: {}. ' +
refer_to_faq_msg(44),
argv.model_name,
str(e)
) from e
graph.__setattr__('name', argv.model_name)
# 'layout' parameter change may cause an issue in EltwiseInputReshape replacer
# and convert_nhwc_to_nchw(graph)
graph.graph['layout'] = 'NCHW' if argv.disable_nhwc_to_nchw else 'NHWC'
graph.graph['fw'] = 'tf'
graph.graph['variables_values'] = variables_values
del variables_values
used_tensors = restore_edges(graph, get_tf_edges)
# Tensor names information corresponding to a node is stored on outgoing edges.
# As output nodes do not have outgoing edges, fake outputs are required. In the following code
# for each output Identity node is added, and tensor name for the output is kept
# on (output, fake output) edge. After Result nodes adding transformation fake outputs
# are deleted from graph.
add_outputs_identity(
graph, graph.nodes - used_tensors,
lambda g, output, fake_node_name: g.add_edges_from(
[create_tf_edge(output, fake_node_name, 0)]))
remove_control_dependency_inputs(graph)
graph.check_empty_graph(
'protobuf2nx. It may happen due to problems with loaded model')
extract_node_attrs(
graph, lambda node: tf_op_extractor(
node, check_for_duplicates(tf_op_extractors)))
# try to detect layout from the nodes of the graph. If there are no convolution nodes in N(D)HWC layout then we
# consider that the graph is in NCHW layout and no layout conversion should be performed
if not argv.disable_nhwc_to_nchw and not argv.silent and not graph_or_sub_graph_has_nhwc_ops(
graph):
log.error(
'The TensorFlow model does not contain Convolution operations with N(D)HWC layout. Most likely '
'the model should be converted using additional "--disable_nhwc_to_nchw" command line parameter '
'which disables model layout conversion inside the Model Optimizer.',
extra={'is_warning': True})
send_op_names_info(framework, graph)
send_shapes_info(framework, graph)
