def automatic_model_search(self, network_info):
model = Path(network_info['model'])
if model.is_dir():
is_blob = network_info.get('_model_is_blob')
if is_blob:
model_list = list(model.glob('*{}.blob'.format(self.default_model_suffix)))
if not model_list:
model_list = list(model.glob('*.blob'))
else:
model_list = list(model.glob('*{}.xml'.format(self.default_model_suffix)))
blob_list = list(model.glob('*{}.blob'.format(self.default_model_suffix)))
if not model_list and not blob_list:
model_list = list(model.glob('*.xml'.format(self.default_model_suffix)))
blob_list = list(model.glob('*.blob'.format(self.default_model_suffix)))
if not model_list:
model_list = blob_list
if not model_list:
raise ConfigError('Suitable model for {} not found'.format(self.default_model_suffix))
if len(model_list) > 1:
raise ConfigError('Several suitable models for {} found'.format(self.default_model_suffix))
model = model_list[0]
if model.suffix == '.blob':
return model, None
weights = network_info.get('weights', model.parent / model.name.replace('xml', 'bin'))
return model, weights
def auto_model_search(self, network_info):
model = Path(network_info.get('model', ''))
weights = network_info.get('weights')
if model.is_dir():
models_list = list(
Path(model).glob('{}.xml'.format(self.default_model_name)))
if not models_list:
models_list = list(Path(model).glob('*.xml'))
if not models_list:
raise ConfigError('Suitable model description is not detected')
if len(models_list) != 1:
raise ConfigError(
'Several suitable models found, please specify required model'
)
model = models_list[0]
if weights is None or Path(weights).is_dir():
weights_dir = weights or model.parent
weights = Path(weights_dir) / model.name.replace('xml', 'bin')
if not weights.exists():
weights_list = list(weights_dir.glob('*.bin'))
if not weights_list:
raise ConfigError('Suitable weights is not detected')
if len(weights_list) != 1:
raise ConfigError(
'Several suitable weights found, please specify required explicitly'
)
weights = weights_list[0]
weights = Path(weights)
return model, weights
def build_stages(models_info, preprocessors_config, launcher, model_args):
def merge_preprocessing(model_specific, common_preprocessing):
if model_specific:
model_specific.extend(common_preprocessing)
return model_specific
return common_preprocessing
required_stages = ['pnet']
stages_mapping = OrderedDict([('pnet', {
'caffe': CaffeProposalStage,
'dlsdk': DLSDKProposalStage,
'dummy': DummyProposalStage
}), ('rnet', {
'caffe': CaffeRefineStage,
'dlsdk': DLSDKRefineStage
}), ('onet', {
'caffe': CaffeOutputStage,
'dlsdk': DLSDKOutputStage
})])
framework = launcher.config['framework']
stages = []
for stage_name, stage_classes in stages_mapping.items():
if stage_name not in models_info:
if stage_name not in required_stages:
continue
else:
raise ConfigError(
'{} required for evaluation'.format(stage_name))
model_config = models_info[stage_name]
if 'predictions' in model_config and not model_config.get(
'store_predictions', False):
stage_framework = 'dummy'
else:
stage_framework = framework
if not contains_any(model_config, ['model', 'caffe_model'
]) and stage_framework != 'dummy':
if model_args:
model_config['model'] = model_args[
len(stages) if len(model_args) > 1 else 0]
stage = stage_classes.get(stage_framework)
if not stage_classes:
raise ConfigError('{} stage does not support {} framework'.format(
stage_name, stage_framework))
stage_preprocess = merge_preprocessing(
models_info[stage_name].get('preprocessing', []),
preprocessors_config)
preprocessor = PreprocessingExecutor(stage_preprocess)
stages.append(stage(models_info[stage_name], preprocessor, launcher))
if not stages:
raise ConfigError(
'please provide information about MTCNN pipeline stages')
return stages
def from_configs(cls, config):
dataset_config = config['datasets'][0]
dataset = Dataset(dataset_config)
data_reader_config = dataset_config.get('reader', 'opencv_imread')
data_source = dataset_config['data_source']
if isinstance(data_reader_config, str):
data_reader_type = data_reader_config
data_reader_config = None
elif isinstance(data_reader_config, dict):
data_reader_type = data_reader_config['type']
else:
raise ConfigError('reader should be dict or string')
if data_reader_type in REQUIRES_ANNOTATIONS:
data_source = dataset.annotation
data_reader = BaseReader.provide(data_reader_type, data_source,
data_reader_config)
models_info = config['network_info']
launcher_config = config['launchers'][0]
launcher = create_launcher(launcher_config, delayed_model_loading=True)
preprocessors_config = dataset_config.get('preprocessing', [])
stages = build_stages(models_info, preprocessors_config, launcher)
metrics_executor = MetricsExecutor(dataset_config['metrics'], dataset)
postprocessing = PostprocessingExecutor(
dataset_config['postprocessing'])
return cls(dataset, data_reader, stages, postprocessing,
metrics_executor)
def __init__(self, model_info, preprocessor, *args, **kwargs):
super().__init__(model_info, preprocessor)
self._index = 0
if 'predictions' not in self.model_info:
raise ConfigError('predictions_file is not found')
self._predictions = read_pickle(self.model_info['predictions'])
self.iterator = 0
def __init__(self, network_info, launcher, models_args, is_blob=None):
detector = network_info.get('detector', {})
recognizer_encoder = network_info.get('recognizer_encoder', {})
recognizer_decoder = network_info.get('recognizer_decoder', {})
if 'model' not in detector:
detector['model'] = models_args[0]
detector['_model_is_blob'] = is_blob
if 'model' not in recognizer_encoder:
recognizer_encoder['model'] = models_args[1 if len(models_args) > 1 else 0]
recognizer_encoder['_model_is_blob'] = is_blob
if 'model' not in recognizer_decoder:
recognizer_decoder['model'] = models_args[2 if len(models_args) > 2 else 0]
recognizer_decoder['_model_is_blob'] = is_blob
network_info.update({
'detector': detector,
'recognizer_encoder': recognizer_encoder,
'recognizer_decoder': recognizer_decoder
})
if not contains_all(network_info, ['detector', 'recognizer_encoder', 'recognizer_decoder']):
raise ConfigError('network_info should contains detector, encoder and decoder fields')
self.detector = create_detector(network_info['detector'], launcher)
self.recognizer_encoder = create_recognizer(network_info['recognizer_encoder'], launcher, 'encoder')
self.recognizer_decoder = create_recognizer(network_info['recognizer_decoder'], launcher, 'decoder')
self.recognizer_decoder_inputs = network_info['recognizer_decoder_inputs']
self.recognizer_decoder_outputs = network_info['recognizer_decoder_outputs']
self.max_seq_len = int(network_info['max_seq_len'])
self.adapter = create_adapter(network_info['adapter'])
self.alphabet = network_info['alphabet']
self.sos_index = int(network_info['sos_index'])
self.eos_index = int(network_info['eos_index'])
def auto_model_search(network_info):
model = Path(network_info['model'])
is_blob = network_info.get('_model_is_blob')
if model.is_dir():
if is_blob:
model_list = list(model.glob('*.blob'))
else:
model_list = list(model.glob('*.xml'))
if not model_list and is_blob is None:
model_list = list(model.glob('*.blob'))
if not model_list:
raise ConfigError('Suitable model not found')
if len(model_list) > 1:
raise ConfigError('Several suitable models found')
model = model_list[0]
if model.suffix == '.blob':
return model, None
weights = network_info.get('weights', model.parent / model.name.replace('xml', 'bin'))
return model, weights
def from_configs(cls, config):
dataset_config = config['datasets'][0]
dataset = Dataset(dataset_config)
data_reader_config = dataset_config.get('reader', 'opencv_imread')
data_source = dataset_config['data_source']
if isinstance(data_reader_config, str):
reader = BaseReader.provide(data_reader_config, data_source)
elif isinstance(data_reader_config, dict):
reader = BaseReader.provide(data_reader_config['type'], data_source, data_reader_config)
else:
raise ConfigError('reader should be dict or string')
preprocessing = PreprocessingExecutor(dataset_config.get('preprocessing', []), dataset.name)
metrics_executor = MetricsExecutor(dataset_config['metrics'], dataset)
launcher_settings = config['launchers'][0]
supported_frameworks = ['dlsdk']
if not launcher_settings['framework'] in supported_frameworks:
raise ConfigError('{} framework not supported'.format(launcher_settings['framework']))
launcher = create_launcher(launcher_settings, delayed_model_loading=True)
network_info = config.get('network_info', {})
colorization_network = network_info.get('colorization_network', {})
verification_network = network_info.get('verification_network', {})
model_args = config.get('_models', [])
models_is_blob = config.get('_model_is_blob')
if 'model' not in colorization_network and model_args:
colorization_network['model'] = model_args[0]
colorization_network['_model_is_blob'] = models_is_blob
if 'model' not in verification_network and model_args:
verification_network['model'] = model_args[1 if len(model_args) > 1 else 0]
verification_network['_model_is_blob'] = models_is_blob
network_info.update({
'colorization_network': colorization_network,
'verification_network': verification_network
})
if not contains_all(network_info, ['colorization_network', 'verification_network']):
raise ConfigError('configuration for colorization_network/verification_network does not exist')
test_model = ColorizationTestModel(network_info['colorization_network'], launcher)
check_model = ColorizationCheckModel(network_info['verification_network'], launcher)
return cls(dataset, reader, preprocessing, metrics_executor, launcher, test_model, check_model)
def __init__(self, network_info, launcher):
super().__init__(network_info, launcher)
if not contains_all(network_info, ['encoder', 'decoder']):
raise ConfigError(
'network_info should contains encoder and decoder fields')
self.num_processing_frames = network_info['decoder'].get(
'num_processing_frames', 16)
self.processing_frames_buffer = []
self.encoder = create_encoder(network_info['encoder'], launcher)
self.decoder = create_decoder(network_info['decoder'], launcher)
self.store_encoder_predictions = network_info['encoder'].get(
'store_predictions', False)
self._encoder_predictions = [] if self.store_encoder_predictions else None
def from_configs(cls, config):
dataset_config = config['datasets'][0]
dataset = Dataset(dataset_config)
data_reader_config = dataset_config.get('reader', 'opencv_imread')
data_source = dataset_config['data_source']
if isinstance(data_reader_config, str):
reader = BaseReader.provide(data_reader_config, data_source)
elif isinstance(data_reader_config, dict):
reader = BaseReader.provide(data_reader_config['type'], data_source, data_reader_config)
else:
raise ConfigError('reader should be dict or string')
preprocessing = PreprocessingExecutor(dataset_config.get('preprocessing', []), dataset.name)
metrics_executor = MetricsExecutor(dataset_config['metrics'], dataset)
launcher = create_launcher(config['launchers'][0], delayed_model_loading=True)
model = SequentialModel(
config.get('network_info', {}), launcher, config.get('_models', []), config.get('_model_is_blob')
)
return cls(dataset, reader, preprocessing, metrics_executor, launcher, model)
def __init__(self, network_info, launcher):
super().__init__(network_info, launcher)
if not contains_all(
network_info,
['detector', 'recognizer_encoder', 'recognizer_decoder']):
raise ConfigError(
'network_info should contains detector, encoder and decoder fields'
)
self.detector = create_detector(network_info['detector'], launcher)
self.recognizer_encoder = create_recognizer(
network_info['recognizer_encoder'], launcher)
self.recognizer_decoder = create_recognizer(
network_info['recognizer_decoder'], launcher)
self.recognizer_decoder_inputs = network_info[
'recognizer_decoder_inputs']
self.recognizer_decoder_outputs = network_info[
'recognizer_decoder_outputs']
self.max_seq_len = int(network_info['max_seq_len'])
self.adapter = create_adapter(network_info['adapter'])
self.alphabet = network_info['alphabet']
self.sos_index = int(network_info['sos_index'])
self.eos_index = int(network_info['eos_index'])
def __init__(self, network_info, launcher):
super().__init__(network_info, launcher)
if 'predictions' not in network_info:
raise ConfigError('predictions_file is not found')
self._predictions = read_pickle(network_info['predictions'])
self.iterator = 0