Barlow Twins
Example implementation of the Barlow Twins architecture.
This example can be run from the command line with:
python lightly/examples/pytorch/barlowtwins.py
# Note: The model and training settings do not follow the reference settings
# from the paper. The settings are chosen such that the example can easily be
# run on a small dataset with a single GPU.
import torch
import torchvision
from torch import nn
from lightly.data import LightlyDataset
from lightly.data.multi_view_collate import MultiViewCollate
from lightly.loss import BarlowTwinsLoss
from lightly.models.modules import BarlowTwinsProjectionHead
from lightly.transforms.simclr_transform import SimCLRTransform
class BarlowTwins(nn.Module):
def __init__(self, backbone):
super().__init__()
self.backbone = backbone
self.projection_head = BarlowTwinsProjectionHead(512, 2048, 2048)
def forward(self, x):
x = self.backbone(x).flatten(start_dim=1)
z = self.projection_head(x)
return z
resnet = torchvision.models.resnet18()
backbone = nn.Sequential(*list(resnet.children())[:-1])
model = BarlowTwins(backbone)
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
cifar10 = torchvision.datasets.CIFAR10("datasets/cifar10", download=True)
transform = SimCLRTransform(input_size=32)
dataset = LightlyDataset.from_torch_dataset(cifar10, transform=transform)
# or create a dataset from a folder containing images or videos:
# dataset = LightlyDataset("path/to/folder")
collate_fn = MultiViewCollate()
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=256,
collate_fn=collate_fn,
shuffle=True,
drop_last=True,
num_workers=8,
)
criterion = BarlowTwinsLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.06)
print("Starting Training")
for epoch in range(10):
total_loss = 0
for (x0, x1), _, _ in dataloader:
x0 = x0.to(device)
x1 = x1.to(device)
z0 = model(x0)
z1 = model(x1)
loss = criterion(z0, z1)
total_loss += loss.detach()
loss.backward()
optimizer.step()
optimizer.zero_grad()
avg_loss = total_loss / len(dataloader)
print(f"epoch: {epoch:>02}, loss: {avg_loss:.5f}")
This example can be run from the command line with:
python lightly/examples/pytorch_lightning/barlowtwins.py
# Note: The model and training settings do not follow the reference settings
# from the paper. The settings are chosen such that the example can easily be
# run on a small dataset with a single GPU.
import pytorch_lightning as pl
import torch
import torchvision
from torch import nn
from lightly.data import LightlyDataset
from lightly.data.multi_view_collate import MultiViewCollate
from lightly.loss import BarlowTwinsLoss
from lightly.models.modules import BarlowTwinsProjectionHead
from lightly.transforms.simclr_transform import SimCLRTransform
class BarlowTwins(pl.LightningModule):
def __init__(self):
super().__init__()
resnet = torchvision.models.resnet18()
self.backbone = nn.Sequential(*list(resnet.children())[:-1])
self.projection_head = BarlowTwinsProjectionHead(512, 2048, 2048)
self.criterion = BarlowTwinsLoss()
def forward(self, x):
x = self.backbone(x).flatten(start_dim=1)
z = self.projection_head(x)
return z
def training_step(self, batch, batch_index):
(x0, x1), _, _ = batch
z0 = self.forward(x0)
z1 = self.forward(x1)
loss = self.criterion(z0, z1)
return loss
def configure_optimizers(self):
optim = torch.optim.SGD(self.parameters(), lr=0.06)
return optim
model = BarlowTwins()
cifar10 = torchvision.datasets.CIFAR10("datasets/cifar10", download=True)
transform = SimCLRTransform(input_size=32)
dataset = LightlyDataset.from_torch_dataset(cifar10, transform=transform)
# or create a dataset from a folder containing images or videos:
# dataset = LightlyDataset("path/to/folder")
collate_fn = MultiViewCollate()
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=256,
collate_fn=collate_fn,
shuffle=True,
drop_last=True,
num_workers=8,
)
accelerator = "gpu" if torch.cuda.is_available() else "cpu"
trainer = pl.Trainer(max_epochs=10, devices=1, accelerator=accelerator)
trainer.fit(model=model, train_dataloaders=dataloader)
This example runs on multiple gpus using Distributed Data Parallel (DDP) training with Pytorch Lightning. At least one GPU must be available on the system. The example can be run from the command line with:
python lightly/examples/pytorch_lightning_distributed/barlowtwins.py
The model differs in the following ways from the non-distributed implementation:
Distributed Data Parallel is enabled
Synchronized Batch Norm is used in place of standard Batch Norm
Features are gathered from all GPUs before the loss is calculated
Note that Synchronized Batch Norm and feature gathering are optional and the model can also be trained without them. Without Synchronized Batch Norm and feature gathering the batch norm and loss for each GPU are only calculated based on the features on that specific GPU.
# Note: The model and training settings do not follow the reference settings
# from the paper. The settings are chosen such that the example can easily be
# run on a small dataset with a single GPU.
import pytorch_lightning as pl
import torch
import torchvision
from torch import nn
from lightly.data import LightlyDataset
from lightly.data.multi_view_collate import MultiViewCollate
from lightly.loss import BarlowTwinsLoss
from lightly.models.modules import BarlowTwinsProjectionHead
from lightly.transforms.simclr_transform import SimCLRTransform
class BarlowTwins(pl.LightningModule):
def __init__(self):
super().__init__()
resnet = torchvision.models.resnet18()
self.backbone = nn.Sequential(*list(resnet.children())[:-1])
self.projection_head = BarlowTwinsProjectionHead(512, 2048, 2048)
# enable gather_distributed to gather features from all gpus
# before calculating the loss
self.criterion = BarlowTwinsLoss(gather_distributed=True)
def forward(self, x):
x = self.backbone(x).flatten(start_dim=1)
z = self.projection_head(x)
return z
def training_step(self, batch, batch_index):
(x0, x1), _, _ = batch
z0 = self.forward(x0)
z1 = self.forward(x1)
loss = self.criterion(z0, z1)
return loss
def configure_optimizers(self):
optim = torch.optim.SGD(self.parameters(), lr=0.06)
return optim
model = BarlowTwins()
cifar10 = torchvision.datasets.CIFAR10("datasets/cifar10", download=True)
transform = SimCLRTransform(input_size=32)
dataset = LightlyDataset.from_torch_dataset(cifar10, transform=transform)
# or create a dataset from a folder containing images or videos:
# dataset = LightlyDataset("path/to/folder")
collate_fn = MultiViewCollate()
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=256,
collate_fn=collate_fn,
shuffle=True,
drop_last=True,
num_workers=8,
)
# Train with DDP and use Synchronized Batch Norm for a more accurate batch norm
# calculation. Distributed sampling is also enabled with replace_sampler_ddp=True.
trainer = pl.Trainer(
max_epochs=10,
devices="auto",
accelerator="gpu",
strategy="ddp",
sync_batchnorm=True,
use_distributed_sampler=True, # or replace_sampler_ddp=True for PyTorch Lightning <2.0
)
trainer.fit(model=model, train_dataloaders=dataloader)