Reproduce “EmoReward: Self-Evolving Reward Calibration for Emotional Video Generation RL” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mingzhe Zheng, Yue Wu, Jing Tan, Chang Zou, Dengyang Jiang, Zeyue Tian, Xuanhua He, Ryan Xu and 14 more
EmoReward calibrates emotional video rewards by gating cues that are usable in each generated clip and using evidence from a completed run to adjust their influence in the next. On HunyuanVideo 1.5, SAGE-GRPO with EmoReward scored −0.2624 ± 0.0600 on VideoAlign Overall, versus −0.8269 ± 0.0532 alone, across 302 held-out prompts; VideoAlign is also an optimization target, not independent proof of emotional fidelity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.