def train(cls,
object_id,
session_ids,
observation_ids,
submethod,
pipeline_params,
db_params,
args=None):
'''
Returns a training plasm, that will be executed exactly once.
:param object_id: The object id to train up.
:param session_ids: A list of session ids that this model should be based on.
:param observation_ids: A list of observation ids that will be dealt to the incremental model builder.
:param submethod: A dictionary of discriminative parameters that will be used to initialize the
training pipeline.
:param pipeline_params: A dictionary of non-discriminative parameters that will be used to initialize the
training pipeline.
:param db_params: A DB parameters object that specifies where to save the model to.
:param args: General command line args, for things like visualize or what have you.
:returns: A plasm, only execute once please.
'''
from ecto_object_recognition.object_recognition_db import ModelWriter
#todo make this depend on the pipeline specification or something...
dealer = ObservationDealer(db_params=db_params,
observation_ids=observation_ids)
pipeline = cls()
incremental_model_builder = pipeline.incremental_model_builder(
submethod, pipeline_params, args)
model_builder = ModelBuilder(
source=dealer,
incremental_model_builder=incremental_model_builder,
niter=0,
) #execute until a quit condition occurs.
post_process = pipeline.post_processor(submethod, pipeline_params,
args)
plasm = ecto.Plasm()
# Connect the model builder to the source
for key in set(model_builder.outputs.keys()).intersection(
post_process.inputs.keys()):
plasm.connect(model_builder[key] >> post_process[key])
writer = ModelWriter(
db_params=db_params,
object_id=object_id,
session_ids=list_to_cpp_json_str(session_ids),
method=cls.type_name(),
json_submethod=dict_to_cpp_json_str(submethod),
json_parameters=dict_to_cpp_json_str(pipeline_params),
)
plasm.connect(post_process["db_document"] >> writer["db_document"])
return plasm
def configure(self, p, _i, _o):
feature_params = self._parameters.get("feature", False)
if not feature_params:
raise RuntimeError("You must supply feature_descriptor parameters for TOD.")
# merge it with the subtype
feature_descriptor_params = { 'feature': feature_params, 'descriptor': self._parameters.get('descriptor', {}) }
from object_recognition.tod import merge_dict
feature_descriptor_params = merge_dict(feature_descriptor_params, self._submethod)
self.feature_descriptor = FeatureDescriptor(json_params=json_helper.dict_to_cpp_json_str(feature_descriptor_params))
self.descriptor_matcher = tod_detection.DescriptorMatcher("Matcher",
search_json_params=json_helper.dict_to_cpp_json_str(self._parameters['search']),
model_documents=self._model_documents)
self.message_cvt = ecto_ros.Mat2Image()
guess_params = self._parameters['guess'].copy()
guess_params['visualize'] = self._visualize
self.guess_generator = tod_detection.GuessGenerator("Guess Gen", **guess_params)
def train(cls, object_id, session_ids, observation_ids, submethod, pipeline_params, db_params, args=None):
'''
Returns a training plasm, that will be executed exactly once.
:param object_id: The object id to train up.
:param session_ids: A list of session ids that this model should be based on.
:param observation_ids: A list of observation ids that will be dealt to the incremental model builder.
:param submethod: A dictionary of discriminative parameters that will be used to initialize the
training pipeline.
:param pipeline_params: A dictionary of non-discriminative parameters that will be used to initialize the
training pipeline.
:param db_params: A DB parameters object that specifies where to save the model to.
:param args: General command line args, for things like visualize or what have you.
:returns: A plasm, only execute once please.
'''
from ecto_object_recognition.object_recognition_db import ModelWriter
#todo make this depend on the pipeline specification or something...
dealer = ObservationDealer(db_params=db_params, observation_ids=observation_ids)
pipeline = cls()
incremental_model_builder = pipeline.incremental_model_builder(submethod, pipeline_params, args)
model_builder = ModelBuilder(source=dealer,
incremental_model_builder=incremental_model_builder,
niter=0,
) #execute until a quit condition occurs.
post_process = pipeline.post_processor(submethod, pipeline_params, args)
plasm = ecto.Plasm()
# Connect the model builder to the source
for key in set(model_builder.outputs.keys()).intersection(post_process.inputs.keys()):
plasm.connect(model_builder[key] >> post_process[key])
writer = ModelWriter(db_params=db_params,
object_id=object_id,
session_ids=list_to_cpp_json_str(session_ids),
method=cls.type_name(),
json_submethod=dict_to_cpp_json_str(submethod),
json_parameters=dict_to_cpp_json_str(pipeline_params),
)
plasm.connect(post_process["db_document"] >> writer["db_document"])
return plasm
def db_models(self, object_ids, submethod, db_params):
return DbModels(db_params, object_ids, self.type_name(), json_helper.dict_to_cpp_json_str(submethod))
def db_models(self, object_ids, submethod, db_params):
return DbModels(db_params, object_ids, self.type_name(),
json_helper.dict_to_cpp_json_str(submethod))
