def analyze(self, graph: Graph):
tf_1_names = ['image_tensor', 'detection_classes', 'detection_boxes', 'detection_scores',
('Preprocessor', 'map')]
tf_1_cond = all([graph_contains_scope(graph, scope) for scope in tf_1_names])
tf_2_names = ['input_tensor', 'output_control_node', 'Identity_', ('Preprocessor', 'map')]
tf_2_cond = all([graph_contains_scope(graph, scope) for scope in tf_2_names])
if not tf_1_cond and not tf_2_cond:
log.debug('The model does not contain nodes that must exist in the TF OD API models')
return None, None
for flavor, scopes_tuple in self.model_scopes.items():
for scopes in scopes_tuple:
if all([graph_contains_scope(graph, scope) for scope in scopes]):
result = dict()
result['flavor'] = flavor
result['mandatory_parameters'] = {'tensorflow_use_custom_operations_config':
files_by_pattern(get_mo_root_dir() + '/openvino/tools/mo/front/tf',
__class__.file_patterns[flavor],
add_prefix=True),
'tensorflow_object_detection_api_pipeline_config': None,
}
message = "Your model looks like TensorFlow Object Detection API Model.\n" \
"Check if all parameters are specified:\n" \
"\t--transformations_config\n" \
"\t--tensorflow_object_detection_api_pipeline_config\n" \
"\t--input_shape (optional)\n" \
"\t--reverse_input_channels (if you convert a model to use with the Inference Engine sample applications)\n" \
"Detailed information about conversion of this model can be found at\n" \
"https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_Object_Detection_API_Models.html"
return {'model_type': {'TF_OD_API': result}}, message
return None, None
def generate_mo_version():
"""
Function generates version like in cmake
custom_{branch_name}_{commit_hash}
"""
try:
mo_dir = get_mo_root_dir()
branch_name = subprocess.check_output(["git", "rev-parse", "--abbrev-ref", "HEAD"], cwd=mo_dir).strip().decode()
commit_hash = subprocess.check_output(["git", "rev-parse", "HEAD"], cwd=mo_dir).strip().decode()
return "{}.custom_{}_{}".format(mo_major_and_minor_version, branch_name, commit_hash)
except Exception as e:
return "unknown version"
class TestSchema(UnitTestWithMockedTelemetry):
base_dir = get_mo_root_dir()
schema_file = os.path.join(base_dir, 'mo', 'utils', 'schema.json')
transformation_configs = get_json_configs(base_dir)
test_json1 = '[{"id": "", "match_kind": "general", "custom_attributes": {}}]'
test_json2 = '[{"id": "someid", "match_kind": "abc", "custom_attributes": {}}]'
@generate(*transformation_configs)
def test_schema_file(self, transformation_config):
self.assertTrue(load_and_validate_json_config(transformation_config))
def test_schema_id_empty(self):
self.assertRaises(Error, load_and_validate_json_config, self.test_json1)
def test_schema_match_kind_wrong(self):
self.assertRaises(Error, load_and_validate_json_config, self.test_json2)
def restore_graph_from_ir(path_to_xml: str, path_to_bin: str = None) -> (Graph, dict):
"""
Function to make valid graph and metadata for MO back stage from IR.
:param path_to_xml:
:param path_to_bin:
:return: (restored graph, meta data)
"""
ir = IREngine(path_to_xml, path_to_bin)
assert ir.graph.graph.get('ir_version') >= 10, 'IR version {} is not supported, ' \
'please generate actual IR for your model and use it.'.format(ir.graph.graph.get('ir_version'))
path = get_mo_root_dir()
collect_ops(path)
collect_extenders(path)
# Create a new copy of graph with correct attributes (shape & type infer, backend attrs etc.)
new_graph = copy_graph_with_ops(ir.graph)
return new_graph, copy(ir.meta_data)
