SimSiam

Example implementation of the SimSiam architecture.

Reference:

Exploring Simple Siamese Representation Learning, 2020

Tutorials:

Tutorial 4: Train SimSiam on Satellite Images

This example can be run from the command line with:

python lightly/examples/pytorch/simsiam.py
import torch
from torch import nn
import torchvision

from lightly.data import LightlyDataset
from lightly.data import SimCLRCollateFunction
from lightly.loss import NegativeCosineSimilarity
from lightly.models.modules import SimSiamProjectionHead
from lightly.models.modules import SimSiamPredictionHead


class SimSiam(nn.Module):
    def __init__(self, backbone):
        super().__init__()
        self.backbone = backbone
        self.projection_head = SimSiamProjectionHead(512, 512, 128)
        self.prediction_head = SimSiamPredictionHead(128, 64, 128)

    def forward(self, x):
        f = self.backbone(x).flatten(start_dim=1)
        z = self.projection_head(f)
        p = self.prediction_head(z)
        z = z.detach()
        return z, p


resnet = torchvision.models.resnet18()
backbone = nn.Sequential(*list(resnet.children())[:-1])
model = SimSiam(backbone)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

cifar10 = torchvision.datasets.CIFAR10("datasets/cifar10", download=True)
dataset = LightlyDataset.from_torch_dataset(cifar10)
# or create a dataset from a folder containing images or videos:
# dataset = LightlyDataset("path/to/folder")

collate_fn = SimCLRCollateFunction(input_size=32)

dataloader = torch.utils.data.DataLoader(
    dataset,
    batch_size=256,
    collate_fn=collate_fn,
    shuffle=True,
    drop_last=True,
    num_workers=8,
)

criterion = NegativeCosineSimilarity()
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, p0 = model(x0)
        z1, p1 = model(x1)
        loss = 0.5 * (criterion(z0, p1) + criterion(z1, p0))
        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}")