RewardVerse: Rubric-Guided Policy Optimization for Video Reward Modeling
Reinforcement learning (RL) is vital for optimizing video generation models, with a robust reward model (RM) serving as the cornerstone. However, existing video reward models often produce unstable scalar scores because they directly map complex, subjective video quality into a single score without explicit evaluation criteria. This leads to scalar drift, where the scoring scal
Reading in Plain English mode — toggle below to switch
Reinforcement learning (RL) is vital for optimizing video generation models, with a robust reward model (RM) serving as the cornerstone. However, existing video reward models often produce unstable scalar scores because they directly map complex, subjective video quality into a single score without explicit evaluation criteria. This leads to scalar drift, where the scoring scal
This work addresses a critical challenge in video generation: stable and reliable reward modeling for RL. By introducing a rubric-based approach, it offers a more robust method for evaluating complex video quality, which is essential for developing high-quality and controllable video generation models.