def __init__(self, config):
super().__init__()
self.config = config
self.batch_size = config.optim_params.batch_size
self.encoder, self.viewmaker, self.system, self.pretrain_config = self.load_pretrained_model()
resnet = self.pretrain_config.model_params.resnet_version
if resnet == 'resnet18':
if self.config.model_params.use_prepool:
if self.pretrain_config.model_params.resnet_small:
num_features = 512 * 4 * 4
else:
num_features = 512 * 7 * 7
else:
num_features = 512
else:
raise Exception(f'resnet {resnet} not supported.')
self.train_dataset, self.val_dataset = datasets.get_image_datasets(
config.data_params.dataset,
default_augmentations=self.pretrain_config.data_params.default_augmentations or False,
)
if not self.pretrain_config.model_params.resnet_small:
self.encoder = nn.Sequential(*list(self.encoder.children())[:-1]) # keep pooling layer
self.encoder = self.encoder.eval()
self.viewmaker = self.viewmaker.eval()
# linear evaluation freezes pretrained weights
utils.frozen_params(self.encoder)
utils.frozen_params(self.viewmaker)
self.num_features = num_features
self.model = self.create_model()
def __init__(self, config):
super(TransferViewMakerSystem, self).__init__()
self.config = config
self.batch_size = config.optim_params.batch_size
self.encoder, self.pretrain_config = self.load_pretrained_model()
resnet = self.pretrain_config.model_params.resnet_version
if resnet == 'resnet18':
if self.config.model_params.use_prepool:
if self.pretrain_config.model_params.resnet_small:
num_features = 512 * 4 * 4
else:
num_features = 512 * 7 * 7
else:
num_features = 512
else:
raise Exception(f'resnet {resnet} not supported.')
if not self.pretrain_config.model_params.resnet_small:
self.encoder = nn.Sequential(*list(self.encoder.children())[:-1]) # keep pooling layer
# Freeze encoder for linear evaluation.
self.encoder = self.encoder.eval()
utils.frozen_params(self.encoder)
default_augmentations = self.pretrain_config.data_params.default_augmentations
if self.config.data_params.force_default_views or default_augmentations == DotMap():
default_augmentations = 'all'
self.train_dataset, self.val_dataset = datasets.get_image_datasets(
config.data_params.dataset,
default_augmentations=default_augmentations,
)
self.num_features = num_features
self.model = self.create_model()
def __init__(self, config):
super().__init__()
self.config = config
self.batch_size = config.optim_params.batch_size
self.encoder, self.system, self.pretrain_config = self.load_pretrained_model(
)
resnet = self.pretrain_config.model_params.resnet_version
if resnet == 'resnet18':
if self.config.model_params.use_prepool:
if self.pretrain_config.model_params.resnet_small:
num_features = 512 * 4 * 4
else:
num_features = 512 * 2 * 2
else:
num_features = 512
elif resnet == 'resnet50':
num_features = 2048
else:
raise Exception(f'resnet {resnet} not supported.')
if not self.pretrain_config.model_params.resnet_small:
raise NotImplementedError()
if not self.config.optim_params.supervised:
self.encoder = self.encoder.eval()
frozen_params(self.encoder)
self.num_features = num_features
self.train_dataset, self.val_dataset = self.create_datasets()
self.model = self.create_model()
def __init__(self, config):
super().__init__()
self.config = config
self.device = f'cuda:{config.gpu_device}' if config.cuda else 'cpu'
self.train_dataset, self.val_dataset = datasets.get_datasets(
config.data_params.dataset)
self.encoder = self.load_pretrained_model()
utils.frozen_params(self.encoder)
self.model = self.create_model()
def __init__(self, config):
super().__init__()
self.config = config
self.batch_size = config.optim_params.batch_size
self.encoder, self.viewmaker, self.system, self.pretrain_config = self.load_pretrained_model(
)
resnet = self.pretrain_config.model_params.resnet_version
if resnet == 'resnet18':
if self.config.model_params.use_prepool:
if self.pretrain_config.model_params.resnet_small:
num_features = 512 * 4 * 4
else:
num_features = 512 * 2 * 2
else:
num_features = 512
elif resnet == 'resnet50':
if self.config.model_params.use_prepool:
num_features = 2048 * 4 * 4
else:
num_features = 2048
else:
raise Exception(f'resnet {resnet} not supported.')
if not self.pretrain_config.model_params.resnet_small:
if self.config.model_params.use_prepool:
cut_ix = -2
else:
cut_ix = -1
self.encoder = nn.Sequential(
*list(self.encoder.children())[:cut_ix])
self.encoder = self.encoder.eval()
frozen_params(self.encoder)
frozen_params(self.viewmaker)
self.num_features = num_features
self.train_dataset, self.val_dataset = self.create_datasets()
self.model = self.create_model()