Reproduce “MATRIX: Mask Track Alignment for Interaction-aware Video Generation” 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

Siyoon Jin, Seongchan Kim, Dahyun Chung, Jaeho Lee, Hyunwook Choi, Jisu Nam, Jiyoung Kim, Seungryong Kim

MATRIX uses persistent instance masks to train selected attention pathways to connect prompt words with the right people and objects, and carry those bindings across video frames. It combines grounding and propagation supervision with the usual video training objective. On a 118-pair image-to-video benchmark, it reaches interaction fidelity of 0.593, versus 0.494 for TaVid, in the evaluated setting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.