The major paradigm of weakly supervised video anomaly detection (WSVAD) is treating it as a multiple instance learning (MIL) problem, with only video-level labels available for training. Due to the ...