Reproduce “Motion-Aware Reward Modeling for Instruction-Guided Video Editing” 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
Li Li, Xiangbo Gao, Bo Ni, Samyadeep Basu, Viet Dac Lai, Jing Shi, Seunghyun Yoon, Gang Wu and 6 more
The authors introduce a motion-aware reward for instruction-guided video editing to prevent motion collapse, where optimization favors suppressed movement. It combines video frames with frame-to-frame RGB differences and trains against suppressed and fabricated motion while preserving quality alignment. On VE-Bench, it has 0% motion-ladder capture and 0.57 Spearman correlation with human ratings, within the evaluated benchmark and two tested pipelines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.