It consists of two main stages: first, it employs intramodal contrastive learning in conjunction with the camera-proxy, updates the memory bank using our innovative Difficulty-aware Cluster-based ...