Reproduce “Movies2Scenes: Using Movie Metadata to Learn Scene Representation” 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
Shixing Chen, Chun-Hao Liu, Xiang Hao, Xiaohan Nie, Maxim Arap, Raffay Hamid
Movies2Scenes learns a visual scene representation by using movie-level metadata, including co-watch, genre, or synopsis, to identify related movies and select scene pairs for contrastive learning. On LVU, it reports average improvements of 7.9% on classification tasks and 9.7% on regression tasks; co-watch pairing beats random pairing in tested validation measures. Downstream evaluations still use labeled task data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.