def __init__(self):
self.best_accuracy = 0.0
# all data
self.data_train = CUBDataset.get_data_all(Config.data_root)
self.task_train = CUBDataset(self.data_train, Config.num_way,
Config.num_shot)
self.task_train_loader = DataLoader(self.task_train,
Config.batch_size,
True,
num_workers=Config.num_workers)
# model
self.matching_net = RunnerTool.to_cuda(Config.matching_net)
RunnerTool.to_cuda(self.matching_net.apply(RunnerTool.weights_init))
self.norm = Normalize(2)
# loss
self.loss = RunnerTool.to_cuda(nn.MSELoss())
# optim
self.matching_net_optim = torch.optim.Adam(
self.matching_net.parameters(), lr=Config.learning_rate)
self.matching_net_scheduler = MultiStepLR(self.matching_net_optim,
Config.train_epoch_lr,
gamma=0.5)
self.test_tool = FSLTestTool(self.matching_test,
data_root=Config.data_root,
num_way=Config.num_way,
num_shot=Config.num_shot,
episode_size=Config.episode_size,
test_episode=Config.test_episode,
transform=self.task_train.transform_test)
pass
def __init__(self):
# all data
self.data_train = OmniglotDataset.get_data_all(Config.data_root)
self.task_train = OmniglotDataset(self.data_train, Config.num_way,
Config.num_shot)
self.task_train_loader = DataLoader(self.task_train,
Config.batch_size,
True,
num_workers=Config.num_workers)
# model
self.matching_net = RunnerTool.to_cuda(
MatchingNet(hid_dim=64, z_dim=64))
RunnerTool.to_cuda(self.matching_net.apply(RunnerTool.weights_init))
self.norm = Normalize(2)
# loss
self.loss = RunnerTool.to_cuda(nn.MSELoss())
# optim
self.matching_net_optim = torch.optim.SGD(
self.matching_net.parameters(),
lr=Config.learning_rate,
momentum=0.9,
weight_decay=5e-4)
self.test_tool = FSLTestTool(self.matching_test,
data_root=Config.data_root,
num_way=Config.num_way_test,
num_shot=Config.num_shot,
episode_size=Config.episode_size,
test_episode=Config.test_episode,
transform=self.task_train.transform_test)
pass
def __init__(self, resnet, low_dim=512, modify_head=False):
super().__init__()
self.resnet = resnet(num_classes=low_dim)
self.l2norm = Normalize(2)
if modify_head:
self.resnet.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)
pass
pass
def __init__(self):
self.adjust_learning_rate = Config.adjust_learning_rate
# all data
self.data_train = OmniglotDataset.get_data_all(Config.data_root)
self.task_train = OmniglotDataset(self.data_train, Config.num_way, Config.num_shot)
self.task_train_loader = DataLoader(self.task_train, Config.batch_size, True, num_workers=Config.num_workers)
# IC
self.produce_class = ProduceClass(len(self.data_train), Config.ic_out_dim, Config.ic_ratio)
self.produce_class.init()
self.task_train.set_samples_class(self.produce_class.classes)
# model
self.matching_net = RunnerTool.to_cuda(Config.matching_net)
self.ic_model = RunnerTool.to_cuda(ICResNet(low_dim=Config.ic_out_dim, encoder=Config.ic_net))
self.norm = Normalize(2)
if Config.multi_gpu:
self.matching_net = RunnerTool.to_cuda(nn.DataParallel(self.matching_net))
cudnn.benchmark = True
pass
# RunnerTool.to_cuda(self.matching_net.apply(RunnerTool.weights_init))
# RunnerTool.to_cuda(self.ic_model.apply(RunnerTool.weights_init))
# optim
self.matching_net_optim = torch.optim.SGD(
self.matching_net.parameters(), lr=Config.learning_rate, momentum=0.9, weight_decay=5e-4)
self.ic_model_optim = torch.optim.SGD(
self.ic_model.parameters(), lr=Config.learning_rate, momentum=0.9, weight_decay=5e-4)
# loss
self.ic_loss = RunnerTool.to_cuda(nn.CrossEntropyLoss())
self.fsl_loss = RunnerTool.to_cuda(nn.MSELoss())
# Eval
self.test_tool_fsl = FSLTestTool(self.matching_test, data_root=Config.data_root,
num_way=Config.num_way_test, num_shot=Config.num_shot,
episode_size=Config.episode_size, test_episode=Config.test_episode,
transform=self.task_train.transform_fsl_test)
self.test_tool_ic = ICTestTool(feature_encoder=None, ic_model=self.ic_model,
data_root=Config.data_root, batch_size=Config.batch_size,
num_workers=Config.num_workers, ic_out_dim=Config.ic_out_dim,
transform=self.task_train_loader.dataset.transform_ic_test, k=20)
pass
def __init__(self):
self.best_accuracy = 0.0
self.adjust_learning_rate = Config.adjust_learning_rate
# all data
self.data_train = TieredImageNetDataset.get_data_all(Config.data_root)
self.task_train = TieredImageNetDataset(self.data_train, Config.num_way,
Config.num_shot, load_data=Config.load_data)
self.task_train_loader = DataLoader(self.task_train, Config.batch_size, True, num_workers=Config.num_workers)
# IC
self.produce_class = ProduceClass(len(self.data_train), Config.ic_out_dim, Config.ic_ratio)
self.produce_class.init()
self.task_train.set_samples_class(self.produce_class.classes)
self.task_train.set_samples_feature(self.produce_class.features)
# model
self.matching_net = RunnerTool.to_cuda(Config.matching_net)
self.ic_model = RunnerTool.to_cuda(ICResNet(low_dim=Config.ic_out_dim,
resnet=Config.resnet, modify_head=Config.modify_head))
self.norm = Normalize(2)
RunnerTool.to_cuda(self.matching_net.apply(RunnerTool.weights_init))
RunnerTool.to_cuda(self.ic_model.apply(RunnerTool.weights_init))
# optim
self.matching_net_optim = torch.optim.SGD(
self.matching_net.parameters(), lr=Config.learning_rate, momentum=0.9, weight_decay=5e-4)
self.ic_model_optim = torch.optim.SGD(
self.ic_model.parameters(), lr=Config.learning_rate, momentum=0.9, weight_decay=5e-4)
# loss
self.ic_loss = RunnerTool.to_cuda(nn.CrossEntropyLoss())
self.fsl_loss = RunnerTool.to_cuda(nn.MSELoss())
# Eval
self.test_tool_fsl = FSLTestTool(self.matching_test, data_root=Config.data_root,
num_way=Config.num_way, num_shot=Config.num_shot,
episode_size=Config.episode_size, test_episode=Config.test_episode,
transform=self.task_train.transform_test)
self.test_tool_ic = ICTestTool(feature_encoder=None, ic_model=self.ic_model, data_root=Config.data_root,
transform=self.task_train.transform_test, batch_size=Config.batch_size,
num_workers=Config.num_workers, ic_out_dim=Config.ic_out_dim, k=Config.knn,
load_data=Config.load_data)
pass
def __init__(self, data_train, classes):
# all data
self.data_train = data_train
self.task_train = TieredImageNetFSLDataset(self.data_train, classes,
Config.fsl_num_way,
Config.fsl_num_shot)
self.task_train_loader = DataLoader(self.task_train,
Config.fsl_batch_size,
True,
num_workers=Config.num_workers)
# model
self.matching_net = RunnerTool.to_cuda(Config.fsl_matching_net)
self.matching_net = RunnerTool.to_cuda(
nn.DataParallel(self.matching_net))
cudnn.benchmark = True
RunnerTool.to_cuda(self.matching_net.apply(RunnerTool.weights_init))
self.norm = Normalize(2)
# optim
self.matching_net_optim = torch.optim.Adam(
self.matching_net.parameters(), lr=Config.fsl_learning_rate)
self.matching_net_scheduler = MultiStepLR(self.matching_net_optim,
Config.fsl_lr_schedule,
gamma=1 / 3)
# loss
self.fsl_loss = RunnerTool.to_cuda(nn.MSELoss())
# Eval
self.test_tool_fsl = FSLTestTool(
self.matching_test,
data_root=Config.data_root,
num_way=Config.fsl_num_way,
num_shot=Config.fsl_num_shot,
episode_size=Config.fsl_episode_size,
test_episode=Config.fsl_test_episode,
transform=self.task_train.transform_test)
pass
def __init__(self, encoder, low_dim=512):
super().__init__()
self.encoder = encoder
self.fc = nn.Linear(self.encoder.out_dim, low_dim)
self.l2norm = Normalize(2)
pass