Reproduce “Deep Fragment Embeddings for Bidirectional Image Sentence Mapping” 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

Andrej Karpathy, Armand Joulin, Li Fei-Fei

The authors present a bidirectional image–sentence retrieval model that aligns sentence dependency relations with detected image regions and scene context. It maps fragment types into a shared space, training local correspondences with multiple-instance learning and retrieval with global ranking. On Flickr30K image annotation, Recall@1 reaches 14.2% versus 11.5% for global ranking alone, but this does not establish grounding or superiority everywhere.

Summary by alphaXiv.

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