About MixUp in Contrastive Learning。

[haozhi1817]

If we adopt the mixup strategy during training, how to samples x_i and x_j from the same class are considered as a positive pair ? Maybe we should sample two identical numbers form beta distribution as the "same class"?

[houjunlin]

Thanks for your interest in our work. Contrastive learning is based on raw augmented samples, rather than mixed samples. It works in parallel with mixup.

[haozhi1817]

Thanks for your interest in our work. Contrastive learning is based on raw augmented samples, rather than mixed samples. It works in parallel with mixup.

Thanks for your prompt reply.