Clarification of x0 and x1 in the simsiam.py example

[bryanbocao]

Thanks for the amazing SSL framework.

I have a few clarification questions regarding x0, x1:
Q1: Is x0 the original image (without being transformed) and x1 is the transformed one from the same image?
Q2: Are both x0 and x1 transformed but differently from the same image?

for epoch in range(10):
    total_loss = 0
    for batch in dataloader:
        x0, x1 = batch[0]
        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}")

https://github.com/lightly-ai/lightly/blob/66ad1b40ebf3b53512703c774988211ce283211f/examples/pytorch/simsiam.py#L58C1-L58C1

Thanks!

[philippmwirth]

Hi @bryanbocao, thanks for the question. For SimSiam transforms are used to generate two views of the original image. So x0 and x1 are different transformed views of the same image.

[guarin]

I'll close this for now, please reopen if you have further questions :)