def __init__(self, config: dict, enable_tune: bool = False, **kwargs):
super().__init__(config=config, enable_tune=enable_tune, **kwargs)
exp_params = config['exp_params']
input_shape = get_example_shape(exp_params['data'])
localizer = create_model(**config['model_params'],
input_shape=input_shape)
self.localizer = localizer
def vae1d(config: dict, run_args: dict) -> VAEExperiment:
exp_params = config['exp_params']
c, l = get_example_shape(exp_params['data'])
model = create_model(**config['model_params'], num_samples=l, channels=c)
return VAEExperiment(model,
params=exp_params,
enable_tune=run_args.get('enable_tune', False))
def vae3d(config: dict, run_args: dict) -> VAEExperiment:
exp_params = config['exp_params']
c, d, h, w = get_example_shape(exp_params['data'])
model = create_model(**config['model_params'],
width=w,
height=h,
depth=d,
channels=c)
return VAEExperiment(model, params=exp_params)
def __init__(self, config: dict, enable_tune: bool = False, **kwargs):
super().__init__(config=config, enable_tune=enable_tune, **kwargs)
input_shape = get_example_shape(config['exp_params']['data'])
self.constraint = create_model(**config['constraint_params'],
input_shape=input_shape)
self.constraint.requires_grad = False
self.constraint.eval()
self.model = create_model(**config['model_params'],
input_shape=input_shape)
def classification_embed2d(config: dict, run_args: dict):
base_experiment, _ = experiment_main(
load_config(config['base_experiment']), run_args)
encoder = base_experiment.model.get_encoder()
encoder.requires_grad = False
exp_params = config['exp_params']
c, h, w = get_example_shape(exp_params['data'])
model = create_model(**config['model_params'],
width=w,
height=h,
channels=c,
encoder=encoder)
return ClassificationExperiment(model, params=exp_params)
def __init__(self,
config: dict,
enable_tune: bool = False,
**kwargs) -> None:
super(VAEExperiment, self).__init__(config=config,
enable_tune=enable_tune,
**kwargs)
params = config['exp_params']
c, h, w = get_example_shape(params['data'])
self.model = create_model(**config['model_params'],
width=w,
height=h,
channels=c,
enable_fid='fid_weight' in params,
progressive_growing=len(
params['progressive_growing'])
if 'progressive_growing' in params else 0)
def __init__(self, config: dict, enable_tune: bool = False, **kwargs):
super().__init__(config=config, enable_tune=enable_tune, **kwargs)
self.classifier = create_model(**config['model_params'],
input_shape=get_example_shape(
config['exp_params']['data']))