Reproduce “Beyond Similarity: Foundation Models as an Efficient Backbone for Training-Free Composed Video Retrieval” 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
Dmitry Demidov, Muhammad Zaigham Zaheer, Omkar Thawakar, Abdelrahman Mohamed Shaker, Rao Anwer
CoVRAGE retrieves target videos for source-clip and edit queries by reusing gallery embeddings, then conditionally reranking and expanding candidates, generating target-oriented descriptions, and verifying close leaders. On CoVR-R, it reports 93.43 R@1 versus 55.48 for the strongest other displayed training-free method, Qwen3-VL-72B. On WebVid-CoVR, its 54.58 R@1 trails MoRe’s 63.00; missed targets outside the bounded pool remain unreachable.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.