Reproduce “Advancing Video-Text Pretraining with Multi-View Captions” 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
Fida M. Thoker, Renaud Vandeghen, Karen Sanchez, Marc Van Droogenbroeck, Bernard Ghanem
MVC improves video-text retrieval by generating concise, detailed, refined, and event-focused captions for each video, then representing summary and detailed views separately during training. In a 1M-video ablation, CARE-T zero-shot text-to-video R@1 rises from 48.0 with summaries alone to 54.1 with all views. The evidence supports the combined recipe, not caption accuracy or lower training cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.