def __init__(self, **kwargs):
""" Constructor. Should take either model config, or NetworkX graph.
:param config: model config to init from
:param graph: NetworkX instance to init from
"""
self._models = []
self._prefix_is_applied = False
self._cache = Dict()
if 'config' not in kwargs and 'graph' not in kwargs:
raise TypeError(
'Unable to load model. Invalid keyword arguments. '
'Expected model config (config=) or NetworkX graph (graph=) is expected '
'with optional target_device parameter. Got {}'.format(
kwargs.keys()))
if 'config' in kwargs:
target_device = kwargs[
'target_device'] if 'target_device' in kwargs else 'ANY'
self._from_config(kwargs['config'], target_device)
elif 'graph' in kwargs:
self._from_graph(kwargs['graph'])
else:
raise TypeError(
'Unable to load models. Invalid keyword argument. '
'Either model config (config=) or NetworkX graph (graph=) is expected.'
)
for model in self._models:
ge.add_fullname_for_nodes(model['model'])
def _from_graph(self, graph):
if graph.graph['ir_version'] == 10:
raise AssertionError(
'POT does not support version 10 of IR.'
'Please convert the model with the newer version of OpenVINO '
'or use the POT from OpenVINO 2021.4.2 to work with version 10 of IR.')
ge.add_fullname_for_nodes(graph)
self._models.append({'model': graph})
self._is_cascade = False
def _from_config(self, model_config, target_device='ANY'):
model_config = model_config if isinstance(model_config, Dict) else Dict(model_config)
if model_config.cascade:
for model_dict in model_config.cascade:
model_config_ = model_config.deepcopy()
model_config_.update(model_dict)
self._models.append({'model': load_graph(model_config_, target_device)})
if len(model_config.cascade) > 1:
self._models[-1]['name'] = model_dict.name
self._models[-1]['model'].name = model_dict.name
else:
self._models.append({'model': load_graph(model_config, target_device)})
self.name = model_config.model_name
self._is_cascade = len(self._models) > 1
if self._is_cascade:
self._add_models_prefix()
for model in self._models:
ge.add_fullname_for_nodes(model['model'])
def _from_graph(self, graph):
ge.add_fullname_for_nodes(graph)
self._models.append({'model': graph})
self._is_cascade = False
