Recently, self-supervised contrastive pre-training has become the de facto regime, that allows for efficient downstream fine-tuning. Meanwhile, its fairness issues are barely studied, though they have drawn great attention from the machine learning …