def moc_emit_ir(ngraph_function: Function, argv: argparse.Namespace):
output_dir = argv.output_dir if argv.output_dir != '.' else os.getcwd()
network = function_to_cnn(ngraph_function)
from mo.back.offline_transformations import apply_user_transformations, apply_moc_transformations
apply_user_transformations(network, parse_transform(argv.transform))
apply_moc_transformations(network)
orig_model_name = os.path.normpath(
os.path.join(output_dir, argv.model_name))
network.serialize(orig_model_name + ".xml", orig_model_name + ".bin")
del argv.feManager
# add meta information to IR
append_ir_info(file=orig_model_name,
meta_info=get_meta_info(argv),
mean_data=None,
input_names=None)
print('[ SUCCESS ] Generated IR version {} model.'.format(
get_ir_version(argv)))
print('[ SUCCESS ] XML file: {}.xml'.format(orig_model_name))
print('[ SUCCESS ] BIN file: {}.bin'.format(orig_model_name))
return 0
def emit_ir(graph: Graph, argv: argparse.Namespace):
NormalizeTI().find_and_replace_pattern(graph)
for_graph_and_each_sub_graph_recursively(
graph,
RemoveConstOps().find_and_replace_pattern)
for_graph_and_each_sub_graph_recursively(
graph,
CreateConstNodesReplacement().find_and_replace_pattern)
prepare_emit_ir(
graph=graph,
data_type=graph.graph['cmd_params'].data_type,
output_dir=argv.output_dir,
output_model_name=argv.model_name,
mean_data=graph.graph['mf'] if 'mf' in graph.graph else None,
input_names=graph.graph['input_names']
if 'input_names' in graph.graph else [],
meta_info=get_meta_info(argv))
if not (argv.framework == 'tf'
and argv.tensorflow_custom_operations_config_update):
output_dir = argv.output_dir if argv.output_dir != '.' else os.getcwd()
print('\n[ SUCCESS ] Generated IR version {} model.'.format(
get_ir_version(argv)))
print('[ SUCCESS ] XML file: {}.xml'.format(
os.path.join(output_dir, argv.model_name)))
print('[ SUCCESS ] BIN file: {}.bin'.format(
os.path.join(output_dir, argv.model_name)))
return 0
def unified_pipeline(argv: argparse.Namespace):
graph = Graph(cmd_params=argv,
name=argv.model_name,
ir_version=get_ir_version(argv))
class_registration.apply_replacements(graph, [
class_registration.ClassType.LOADER,
class_registration.ClassType.FRONT_REPLACER,
class_registration.ClassType.MIDDLE_REPLACER,
class_registration.ClassType.BACK_REPLACER
])
return graph
def driver(argv: argparse.Namespace):
caffe_pb2 = loader.import_caffe_pb2(argv.caffe_parser_path)
proto, model = loader.load_caffe_proto_model(caffe_pb2, argv.input_proto,
argv.input_model)
update_extractors_with_extensions(
caffe_type_extractors, argv.disable_omitting_optional if hasattr(
argv, 'disable_omitting_optional') else False,
argv.disable_flattening_optional_params if hasattr(
argv, 'disable_flattening_optional_params') else False)
try:
graph, original_shapes = loader.caffe_pb_to_nx(proto, model)
except ValueError as e:
raise Error(
'Invalid prototxt file: value error {}. ' + refer_to_faq_msg(11),
str(e)) from e
graph.check_empty_graph('load_caffe_proto_model')
graph.__setattr__('proto_path', argv.input_proto)
graph.__setattr__('caffemodel_path', argv.input_model)
graph.__setattr__('name', getattr(proto, 'name', None) or argv.model_name)
graph.graph['layout'] = 'NCHW'
graph.graph['cmd_params'] = argv
graph.graph['fw'] = 'caffe'
graph.graph['original_shapes'] = original_shapes
graph.graph['caffe_pb2'] = caffe_pb2
graph.graph['ir_version'] = get_ir_version(argv)
graph.graph['original_shapes'] = original_shapes
graph.graph['caffe_pb2'] = caffe_pb2
custom_layers_map = custom_layers_mapping.load_layers_xml(argv.k)
custom_layers_mapping.update_extractors(
caffe_type_extractors,
custom_layers_map, argv.disable_omitting_optional if hasattr(
argv, 'disable_omitting_optional') else False,
argv.enable_flattening_nested_params if hasattr(
argv, 'enable_flattening_nested_params') else False)
extract_node_attrs(
graph, lambda node: caffe_extractor(
node, check_for_duplicates(caffe_type_extractors)))
# --------------------------------- LOAD END ------------------------------------------------------
class_registration.apply_replacements(graph, [
class_registration.ClassType.FRONT_REPLACER,
class_registration.ClassType.MIDDLE_REPLACER,
class_registration.ClassType.BACK_REPLACER
])
return graph
def driver(argv: argparse.Namespace):
model_proto = load_onnx_model(argv.input_model)
model_graph = model_proto.graph # pylint: disable=no-member
# print(model_graph)
# assert len(model_graph) == 1, "An ONNX model contains more than 1 graph: unsupported"
log.debug("Number of nodes in graph_def: {}".format(len(model_graph.node)))
log.debug(
"Number of all input ports (not true inputs) in graph_def: {}".format(
len(model_graph.input)))
log.debug("Number of initializers in graph_def: {}".format(
len(model_graph.initializer)))
log.debug("Number of real inputs in graph_def: {}".format(
len(model_graph.input) - len(model_graph.initializer)))
update_extractors_with_extensions(onnx_op_extractors)
try:
graph = protobuf2nx(model_proto)
log.debug("Number of nodes in NX graph: {}".format(
graph.number_of_nodes()))
graph.__setattr__(
'name',
argv.model_name if argv.model_name else model_proto.graph.name) # pylint: disable=no-member
graph.graph['layout'] = 'NCHW'
graph.graph['cmd_params'] = argv
graph.graph['fw'] = 'onnx'
graph.graph[
'feature_dim'] = 1 if graph.graph['layout'] == 'NCHW' else 3
graph.graph['ir_version'] = get_ir_version(argv)
except Exception as e:
raise Error(
'Cannot pre-process ONNX graph after reading from model file "{}". ' \
'File is corrupt or has unsupported format. Details: {}. ' +
refer_to_faq_msg(44),
argv.input_model,
str(e)
) from e
graph.check_empty_graph(
'protobuf2nx. It may happen due to problems with loaded model')
extract_node_attrs(
graph, lambda node: onnx_op_extractor(
node, check_for_duplicates(onnx_op_extractors)))
# --------------------------------- LOAD END ------------------------------------------------------
class_registration.apply_replacements(graph, [
class_registration.ClassType.FRONT_REPLACER,
class_registration.ClassType.MIDDLE_REPLACER,
class_registration.ClassType.BACK_REPLACER
])
return graph
def emit_ir(graph: Graph, argv: argparse.Namespace):
NormalizeTI().find_and_replace_pattern(graph)
for_graph_and_each_sub_graph_recursively(
graph,
RemoveConstOps().find_and_replace_pattern)
for_graph_and_each_sub_graph_recursively(
graph,
CreateConstNodesReplacement().find_and_replace_pattern)
prepare_emit_ir(
graph=graph,
data_type=graph.graph['cmd_params'].data_type,
output_dir=argv.output_dir,
output_model_name=argv.model_name,
mean_data=graph.graph['mf'] if 'mf' in graph.graph else None,
input_names=graph.graph['input_names']
if 'input_names' in graph.graph else [],
meta_info=get_meta_info(argv))
if not (argv.framework == 'tf'
and argv.tensorflow_custom_operations_config_update):
output_dir = argv.output_dir if argv.output_dir != '.' else os.getcwd()
orig_model_name = os.path.normpath(
os.path.join(output_dir, argv.model_name))
# This try-except is additional reinsurance that the IE
# dependency search does not break the MO pipeline
try:
if find_ie_version(silent=True):
path_to_offline_transformations = os.path.join(
os.path.realpath(os.path.dirname(__file__)), 'back',
'offline_transformations.py')
status = subprocess.run([
sys.executable, path_to_offline_transformations,
orig_model_name
],
env=os.environ,
timeout=100)
if status.returncode != 0 and not argv.silent:
print("[ WARNING ] offline_transformations return code {}".
format(status.returncode))
except Exception as e:
# TODO: send error message
pass
print('[ SUCCESS ] Generated IR version {} model.'.format(
get_ir_version(argv)))
print('[ SUCCESS ] XML file: {}.xml'.format(orig_model_name))
print('[ SUCCESS ] BIN file: {}.bin'.format(orig_model_name))
return 0
def driver(argv: argparse.Namespace):
try:
model_nodes, model_params, model_name, iteration_number = load_symbol_def(
argv.input_model, argv.input_symbol, argv.input,
argv.nd_prefix_name, argv.pretrained_model_name,
argv.legacy_mxnet_model)
except (ValueError, mxnet.base.MXNetError) as e:
raise FrameworkError(
'The following error happened while loading mxnet model {}: {}. ' +
refer_to_faq_msg(53), argv.input_model, str(e)) from e
if argv.nd_prefix_name and argv.pretrained_model_name and argv.save_params_from_nd:
save_params_file(model_name, model_params._arg_params,
model_params._aux_params, iteration_number)
update_extractors_with_extensions(mxnet_op_extractors)
graph = symbol2nx(model_nodes, model_params, argv.input)
graph.check_empty_graph(
'symbol2nx. It may happen due to problems with loaded model')
graph.__setattr__('name', argv.model_name)
graph.graph['layout'] = 'NCHW'
graph.graph['cmd_params'] = argv
graph.graph['fw'] = 'mxnet'
graph.graph['feature_dim'] = 1 if graph.graph['layout'] == 'NCHW' else 3
graph.graph['ir_version'] = get_ir_version(argv)
extract_node_attrs(graph, mxnet_op_extractor)
# --------------------------------- LOAD END ------------------------------------------------------
class_registration.apply_replacements(graph, [
class_registration.ClassType.FRONT_REPLACER,
class_registration.ClassType.MIDDLE_REPLACER,
class_registration.ClassType.BACK_REPLACER
])
return graph
def driver(argv):
try:
graph = load_kaldi_model(argv.input_model)
except Exception as e:
raise Error('Model Optimizer is not able to parse Kaldi model {}. '.
format(argv.input_model) + refer_to_faq_msg(91)) from e
graph.check_empty_graph('load_kaldi_nnet_model')
graph.graph['layout'] = 'NCHW'
graph.graph['cmd_params'] = argv
graph.graph['fw'] = 'kaldi'
graph.graph['ir_version'] = get_ir_version(argv)
update_extractors_with_extensions(kaldi_type_extractors)
extract_node_attrs(graph, lambda node: kaldi_extractor(node))
# --------------------------------- LOAD END ------------------------------------------------------
class_registration.apply_replacements(graph, [
class_registration.ClassType.FRONT_REPLACER,
class_registration.ClassType.MIDDLE_REPLACER,
class_registration.ClassType.BACK_REPLACER
])
return graph
def emit_ir(graph: Graph, argv: argparse.Namespace):
NormalizeTI().find_and_replace_pattern(graph)
for_graph_and_each_sub_graph_recursively(
graph,
RemoveConstOps().find_and_replace_pattern)
for_graph_and_each_sub_graph_recursively(
graph,
CreateConstNodesReplacement().find_and_replace_pattern)
if 'feManager' in argv:
del argv.feManager
mean_data = deepcopy(graph.graph['mf']) if 'mf' in graph.graph else None
input_names = deepcopy(
graph.graph['input_names']) if 'input_names' in graph.graph else []
prepare_emit_ir(graph=graph,
data_type=graph.graph['cmd_params'].data_type,
output_dir=argv.output_dir,
output_model_name=argv.model_name,
mean_data=mean_data,
input_names=input_names,
meta_info=get_meta_info(argv),
use_temporary_path=True)
# This graph cleanup is required to avoid double memory consumption
graph.clear()
if not (argv.framework == 'tf'
and argv.tensorflow_custom_operations_config_update):
output_dir = argv.output_dir if argv.output_dir != '.' else os.getcwd()
orig_model_name = os.path.normpath(
os.path.join(output_dir, argv.model_name))
return_code = "not executed"
# This try-except is additional reinsurance that the IE
# dependency search does not break the MO pipeline
try:
if not argv.legacy_ir_generation:
path_to_offline_transformations = os.path.join(
os.path.realpath(os.path.dirname(__file__)), 'back',
'offline_transformations.py')
cmd = [
sys.executable, path_to_offline_transformations,
"--input_model", orig_model_name, "--framework",
argv.framework, "--transform", argv.transform
]
if "compress_fp16" in argv and argv.compress_fp16:
cmd += ["--compress_fp16"]
# restore data_type cmd parameter
argv.data_type = 'FP16'
status = subprocess.run(cmd, env=os.environ)
return_code = status.returncode
except Exception as e:
return_code = "failed"
log.error(e)
message = str(
dict({
"platform": platform.system(),
"mo_version": get_simplified_mo_version(),
"ie_version": get_simplified_ie_version(env=os.environ),
"python_version": sys.version,
"return_code": return_code
}))
t = tm.Telemetry()
t.send_event('mo', 'offline_transformations_status', message)
if return_code != 0:
raise Error("offline transformations step has failed.")
for suf in [".xml", ".bin", ".mapping"]:
# remove existing files
path_to_file = orig_model_name + "_tmp" + suf
if os.path.exists(path_to_file):
os.remove(path_to_file)
# add meta information to IR
append_ir_info(file=orig_model_name,
meta_info=get_meta_info(argv),
mean_data=mean_data,
input_names=input_names)
print('[ SUCCESS ] Generated IR version {} model.'.format(
get_ir_version(argv)))
print('[ SUCCESS ] XML file: {}.xml'.format(orig_model_name))
print('[ SUCCESS ] BIN file: {}.bin'.format(orig_model_name))
return 0
def emit_ir(graph: Graph, argv: argparse.Namespace):
NormalizeTI().find_and_replace_pattern(graph)
for_graph_and_each_sub_graph_recursively(
graph,
RemoveConstOps().find_and_replace_pattern)
for_graph_and_each_sub_graph_recursively(
graph,
CreateConstNodesReplacement().find_and_replace_pattern)
mean_data = deepcopy(graph.graph['mf']) if 'mf' in graph.graph else None
input_names = deepcopy(
graph.graph['input_names']) if 'input_names' in graph.graph else []
# Remove temporary ie_is_available key from argv no to have it in IR
ie_is_available = argv.ie_is_available
del argv.ie_is_available
prepare_emit_ir(graph=graph,
data_type=graph.graph['cmd_params'].data_type,
output_dir=argv.output_dir,
output_model_name=argv.model_name,
mean_data=mean_data,
input_names=input_names,
meta_info=get_meta_info(argv),
use_temporary_path=True)
# This graph cleanup is required to avoid double memory consumption
graph.clear()
if not (argv.framework == 'tf'
and argv.tensorflow_custom_operations_config_update):
output_dir = argv.output_dir if argv.output_dir != '.' else os.getcwd()
orig_model_name = os.path.normpath(
os.path.join(output_dir, argv.model_name))
return_code = "not executed"
# This try-except is additional reinsurance that the IE
# dependency search does not break the MO pipeline
try:
if not argv.legacy_ir_generation and ie_is_available:
path_to_offline_transformations = os.path.join(
os.path.realpath(os.path.dirname(__file__)), 'back',
'offline_transformations.py')
status = subprocess.run([
sys.executable, path_to_offline_transformations,
"--input_model", orig_model_name, "--framework",
argv.framework, "--transform", argv.transform
],
env=os.environ)
return_code = status.returncode
except Exception as e:
return_code = "failed"
log.error(e, extra={'is_warning': True})
message = str(
dict({
"platform": platform.system(),
"mo_version": get_simplified_mo_version(),
"ie_version": get_simplified_ie_version(env=os.environ),
"python_version": sys.version,
"return_code": return_code
}))
t = tm.Telemetry()
t.send_event('mo', 'offline_transformations_status', message)
# if IR wasn't produced by offline_transformations step we need to fallback to IR
# produced by prepare_ir. This IR needs to be renamed from XXX_tmp.xml to XXX.xml
suffixes = [".xml", ".bin", ".mapping"]
if return_code != 0:
if len(argv.transform) != 0:
# Remove temporary IR before throwing exception
for suf in suffixes:
path_to_file = orig_model_name + "_tmp" + suf
if os.path.exists(path_to_file):
os.remove(path_to_file)
raise Error("Failed to apply transformations: {}".format(
argv.transform))
log.error("Using fallback to produce IR.",
extra={'is_warning': True})
for suf in suffixes:
# remove existing files
path_to_file = orig_model_name + suf
if os.path.exists(path_to_file):
os.remove(path_to_file)
# rename tmp IR to original name
os.rename(orig_model_name + "_tmp" + suf,
orig_model_name + suf)
else:
for suf in suffixes:
# remove existing files
path_to_file = orig_model_name + "_tmp" + suf
if os.path.exists(path_to_file):
os.remove(path_to_file)
# add meta information to IR
append_ir_info(file=orig_model_name,
meta_info=get_meta_info(argv),
mean_data=mean_data,
input_names=input_names)
print('[ SUCCESS ] Generated IR version {} model.'.format(
get_ir_version(argv)))
print('[ SUCCESS ] XML file: {}.xml'.format(orig_model_name))
print('[ SUCCESS ] BIN file: {}.bin'.format(orig_model_name))
return 0
def driver(argv: argparse.Namespace):
"""
Convert TF GraphDef object to NetworkX representation.
The resulting graph is still TF-specific and needs normalization passes to be applied.
The specific TF structure assumes each GraphDef node is converted to a single
NetworkX node, node id is an original TF node name, and edges go directly from one op to another op.
"""
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 = 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)
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:
graph = protobuf2nx(graph_def)
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['cmd_params'] = argv
graph.graph['fw'] = 'tf'
graph.graph['ir_version'] = get_ir_version(argv)
graph.graph['variables_values'] = variables_values
del variables_values
graph = restore_edges(graph, get_tf_edges)
graph = remove_control_dependency_inputs(graph)
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.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)))
# --------------------------------- LOAD END ------------------------------------------------------
class_registration.apply_replacements(graph, [
class_registration.ClassType.FRONT_REPLACER,
class_registration.ClassType.MIDDLE_REPLACER,
class_registration.ClassType.BACK_REPLACER
])
return graph
