Reproduce “Maximal Matching Matters: Preventing Representation Collapse for Robust Cross-Modal 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
Hani Alomari, Anushka Sivakumar, Andrew Zhang, Chris Thomas
MaxMatch represents images and captions as sets of vectors, then uses the Hungarian algorithm to choose one-to-one pairings with the greatest total cosine similarity. Global Discriminative and Intra-Set Divergence losses discourage redundant or globally collapsed vectors. On COCO 5K, its full configuration reaches 446.53 RSUM versus 444.10 for Smooth-Chamfer, in that ablation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.