def build_model(inputs, num_classes, is_training, hparams):
"""Constructs the vision model being trained/evaled.
Args:
inputs: input features/images being fed to the image model build built.
num_classes: number of output classes being predicted.
is_training: is the model training or not.
hparams: additional hyperparameters associated with the image model.
Returns:
The logits of the image model.
"""
scopes = setup_arg_scopes(is_training)
with contextlib.nested(*scopes):
if hparams.model_name == 'pyramid_net':
logits = build_shake_drop_model(inputs, num_classes, is_training)
elif hparams.model_name == 'wrn':
logits = build_wrn_model(inputs, num_classes, hparams.wrn_size)
elif hparams.model_name == 'shake_shake':
logits = build_shake_shake_model(inputs, num_classes, hparams,
is_training)
elif hparams.model_name == 'resnet':
logits = build_resnet_model(inputs, num_classes, hparams,
is_training)
else:
raise ValueError("Unknown model name.")
return logits
def build_model(inputs, num_classes, is_training, update_bn, hparams):
"""Constructs the vision model being trained/evaled.
Args:
inputs: input features/images being fed to the image model build built.
num_classes: number of output classes being predicted.
is_training: is the model training or not.
hparams: additional hyperparameters associated with the image model.
Returns:
The logits of the image model.
"""
scopes = setup_arg_scopes(is_training)
with contextlib.nested(*scopes):
if hparams.model_name == "pyramid_net":
logits = build_shake_drop_model(inputs, num_classes, is_training)
elif hparams.model_name == "wrn":
logits = build_wrn_model(inputs, num_classes, hparams.wrn_size,
update_bn)
elif hparams.model_name == "shake_shake":
logits = build_shake_shake_model(inputs, num_classes, hparams,
is_training)
return logits