To mitigate this issue, we propose a Temporal Diversified Self-Contrastive Learning (TDSCL) framework, which guides the model to exploit generalized temporal inconsistencies for face forgery detection ...