def cli_main():
from pl_bolts.callbacks.self_supervised import SSLOnlineEvaluator
from pl_bolts.datamodules import CIFAR10DataModule, STL10DataModule, ImagenetDataModule
from pl_bolts.models.self_supervised.simclr import SimCLRTrainDataTransform, SimCLREvalDataTransform
seed_everything(1234)
parser = ArgumentParser()
# trainer args
parser = pl.Trainer.add_argparse_args(parser)
# model args
parser = BYOL.add_model_specific_args(parser)
args = parser.parse_args()
# pick data
dm = None
# init default datamodule
if args.dataset == 'cifar10':
dm = CIFAR10DataModule.from_argparse_args(args)
dm.train_transforms = SimCLRTrainDataTransform(32)
dm.val_transforms = SimCLREvalDataTransform(32)
args.num_classes = dm.num_classes
elif args.dataset == 'stl10':
dm = STL10DataModule.from_argparse_args(args)
dm.train_dataloader = dm.train_dataloader_mixed
dm.val_dataloader = dm.val_dataloader_mixed
(c, h, w) = dm.size()
dm.train_transforms = SimCLRTrainDataTransform(h)
dm.val_transforms = SimCLREvalDataTransform(h)
args.num_classes = dm.num_classes
elif args.dataset == 'imagenet2012':
dm = ImagenetDataModule.from_argparse_args(args, image_size=196)
(c, h, w) = dm.size()
dm.train_transforms = SimCLRTrainDataTransform(h)
dm.val_transforms = SimCLREvalDataTransform(h)
args.num_classes = dm.num_classes
model = BYOL(**args.__dict__)
def to_device(batch, device):
(x1, x2), y = batch
x1 = x1.to(device)
y = y.to(device)
return x1, y
# finetune in real-time
online_eval = SSLOnlineEvaluator(z_dim=2048, num_classes=dm.num_classes)
online_eval.to_device = to_device
trainer = pl.Trainer.from_argparse_args(args,
max_steps=300000,
callbacks=[online_eval])
trainer.fit(model, dm)
def cli_main():
from pl_bolts.datamodules import CIFAR10DataModule, STL10DataModule
from pl_bolts.datamodules.ssl_imagenet_datamodule import SSLImagenetDataModule
pl.seed_everything(1234)
parser = ArgumentParser()
parser = pl.Trainer.add_argparse_args(parser)
parser = CPCV2.add_model_specific_args(parser)
args = parser.parse_args()
args.online_ft = True
datamodule = None
online_evaluator = SSLOnlineEvaluator()
if args.dataset == 'cifar10':
datamodule = CIFAR10DataModule.from_argparse_args(args)
datamodule.train_transforms = CPCTrainTransformsCIFAR10()
datamodule.val_transforms = CPCEvalTransformsCIFAR10()
args.patch_size = 8
elif args.dataset == 'stl10':
datamodule = STL10DataModule.from_argparse_args(args)
datamodule.train_dataloader = datamodule.train_dataloader_mixed
datamodule.val_dataloader = datamodule.val_dataloader_mixed
datamodule.train_transforms = CPCTrainTransformsSTL10()
datamodule.val_transforms = CPCEvalTransformsSTL10()
args.patch_size = 16
# 16 GB RAM - 64
# 32 GB RAM - 144
args.batch_size = 144
def to_device(batch, device):
(_, _), (x2, y2) = batch
x2 = x2.to(device)
y2 = y2.to(device)
return x2, y2
online_evaluator.to_device = to_device
elif args.dataset == 'imagenet2012':
datamodule = SSLImagenetDataModule.from_argparse_args(args)
datamodule.train_transforms = CPCTrainTransformsImageNet128()
datamodule.val_transforms = CPCEvalTransformsImageNet128()
args.patch_size = 32
model = CPCV2(**vars(args), datamodule=datamodule)
trainer = pl.Trainer.from_argparse_args(args, callbacks=[online_evaluator])
trainer.fit(model)
def cli_main():
from pl_bolts.datamodules import CIFAR10DataModule, STL10DataModule, ImagenetDataModule
parser = ArgumentParser()
# trainer args
parser = pl.Trainer.add_argparse_args(parser)
# model args
parser = SimCLR.add_model_specific_args(parser)
args = parser.parse_args()
# init default datamodule
if args.dataset == 'cifar10':
dm = CIFAR10DataModule.from_argparse_args(args)
dm.train_transforms = SimCLRTrainDataTransform(32)
dm.val_transforms = SimCLREvalDataTransform(32)
args.num_samples = dm.num_samples
elif args.dataset == 'stl10':
dm = STL10DataModule.from_argparse_args(args)
dm.train_dataloader = dm.train_dataloader_mixed
dm.val_dataloader = dm.val_dataloader_mixed
args.num_samples = dm.num_unlabeled_samples
(c, h, w) = dm.size()
dm.train_transforms = SimCLRTrainDataTransform(h)
dm.val_transforms = SimCLREvalDataTransform(h)
elif args.dataset == 'imagenet2012':
dm = ImagenetDataModule.from_argparse_args(args, image_size=196)
(c, h, w) = dm.size()
dm.train_transforms = SimCLRTrainDataTransform(h)
dm.val_transforms = SimCLREvalDataTransform(h)
model = SimCLR(**args.__dict__)
# finetune in real-time
def to_device(batch, device):
(x1, x2), y = batch
x1 = x1.to(device)
y = y.to(device)
return x1, y
online_eval = SSLOnlineEvaluator(z_dim=2048 * 2 * 2,
num_classes=dm.num_classes)
online_eval.to_device = to_device
trainer = pl.Trainer.from_argparse_args(args, callbacks=[online_eval])
trainer.fit(model, dm)
dm = STL10DataModule.from_argparse_args(args)
dm.train_dataloader = dm.train_dataloader_mixed
dm.val_dataloader = dm.val_dataloader_mixed
args.num_samples = dm.num_unlabeled_samples
(c, h, w) = dm.size()
dm.train_transforms = SimCLRTrainDataTransform(h)
dm.val_transforms = SimCLREvalDataTransform(h)
elif args.dataset == 'imagenet2012':
dm = ImagenetDataModule.from_argparse_args(args, image_size=196)
(c, h, w) = dm.size()
dm.train_transforms = SimCLRTrainDataTransform(h)
dm.val_transforms = SimCLREvalDataTransform(h)
model = SimCLR(**args.__dict__)
# finetune in real-time
def to_device(batch, device):
(x1, x2), y = batch
x1 = x1.to(device)
y = y.to(device)
return x1, y
online_eval = SSLOnlineEvaluator(z_dim=2048 * 2 * 2,
num_classes=dm.num_classes)
online_eval.to_device = to_device
trainer = pl.Trainer.from_argparse_args(args, callbacks=[online_eval])
trainer.fit(model, dm)
parser = pl.Trainer.add_argparse_args(parser)
# model args
parser = BYOL.add_model_specific_args(parser)
args = parser.parse_args()
# pick data
datamodule = None
if args.dataset == 'stl10':
datamodule = STL10DataModule.from_argparse_args(args)
datamodule.train_dataloader = datamodule.train_dataloader_mixed
datamodule.val_dataloader = datamodule.val_dataloader_mixed
(c, h, w) = datamodule.size()
datamodule.train_transforms = SimCLRTrainDataTransform(h)
datamodule.val_transforms = SimCLREvalDataTransform(h)
elif args.dataset == 'imagenet2012':
datamodule = ImagenetDataModule.from_argparse_args(args,
image_size=196)
(c, h, w) = datamodule.size()
datamodule.train_transforms = SimCLRTrainDataTransform(h)
datamodule.val_transforms = SimCLREvalDataTransform(h)
model = BYOL(**args.__dict__, datamodule=datamodule)
trainer = pl.Trainer.from_argparse_args(args,
max_steps=10000,
callbacks=[SSLOnlineEvaluator()])
trainer.fit(model)