I got curious while reading the paper. Why do we use just one positive sample?

[babbu3682]

I got curious while reading the paper. Why do we use just one positive sample?

I mean we can generate more positive samples by using strong augmentations.

I think many positive also help representation learning.

Do you already experiment?

[chentingpc]

Using more positive examples would converge faster (in iterations) but at a cost of more compute per iteration. There's a trick for doing it more efficient in SWAV (i.e. multi-crop).