Controllable Multi-label Video Safety Detection via Adaptive Tversky Policy Optimization
Adaptive Tversky Policy Optimization (ATPO) is proposed, a reinforcement learning framework for Multi-label Video Safety Detection (Multi-VSD), which introduces the Adaptive Tversky Reward (ATR), which dynamically adjusts false-positive and false-negative penalties during training to enable controllable precision-recal...
Guang-Yu Yang, Jing-Biao Mei, Ming-Shen Sun et al.
· 0 citations