def __init__(self,
dataset,
train_dataset,
cls,
detname='perfect',
detector_config=None):
self.dataset = dataset
self.train_dataset = train_dataset
self.cls = cls
self.cls_ind = dataset.get_ind(cls)
self.detname = detname
self.tt = TicToc()
# Check if configs exist and look up the correct config for this detname and cls
# TODO: this is inefficient because this file is re-opened for every
# class
loaded_detector_config = None
fname = os.path.join(
train_dataset.config.get_dets_configs_dir(train_dataset),
detname + '.txt')
if os.path.exists(fname):
with open(fname) as f:
configs = json.load(f)
config_name = detname + '_' + cls
if config_name in configs:
loaded_detector_config = configs[config_name]
self.config = loaded_detector_config
if not self.config:
self.config = Detector.DEFAULT_CONFIG
if detector_config:
self.config.update(detector_config)
def __init__(self,
config,
test_dataset,
train_dataset,
weights_dataset_name=None,
**kwargs):
"""
Initialize the DatasetPolicy, getting it ready to run on the whole dataset
or a single image.
- test, train, weights datasets should be:
- val, train, val: for training the weights
- test, trainval, val: for final run
- **kwargs update the default config
"""
self.config = config
# TODO: get rid of doing it this way, be explicit every time
policy_config = copy.copy(DatasetPolicy.default_config)
policy_config.update(kwargs)
self.dataset = test_dataset
self.train_dataset = train_dataset
if not weights_dataset_name:
weights_dataset_name = self.dataset.name
self.weights_dataset_name = weights_dataset_name
self.__dict__.update(policy_config)
print("DatasetPolicy running with:")
pprint(self.__dict__)
self.ev = Evaluation(self.config, self)
self.tt = TicToc()
# Determine inference mode:
# fixed_order, random, oracle policy modes get fixed_order inference
# mode
if re.search('fastinf', self.policy_mode):
self.inference_mode = 'fastinf'
elif self.policy_mode == 'random':
self.inference_mode = 'random'
elif self.policy_mode in ['no_smooth', 'backoff']:
self.inference_mode = self.policy_mode
else:
self.inference_mode = 'fixed_order'
# Determine fastinf model name
self.fastinf_model_name = 'this_is_empty'
if self.inference_mode == 'fastinf':
if self.detectors == ['csc_default']:
self.fastinf_model_name = 'CSC'
elif self.detectors == ['perfect']:
self.fastinf_model_name = 'perfect'
elif self.detectors == ['gist']:
self.fastinf_model_name = 'GIST'
elif self.detectors == ['gist', 'csc_default']:
self.fastinf_model_name = 'GIST_CSC'
else:
raise RuntimeError("""
We don't have Fastinf models for the detector combination you
are running with: %s""" % self.detectors)
# load the actions and the corresponding weights and we're ready to go
self.test_actions = self.init_actions()
self.actions = self.test_actions
self.load_weights()