Time-Contrastive Networks: Self-Supervised Learning from Video

   Abstract