Reproduce “Compositionality with Primitives: A Unified View from Euclidean to Hyperbolic Vision-Language Models” 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

Puru Vaish, Meelad Dashti, Christoph Brune, Pascal Mettes, Tatiana Tommasi, Nicola Strisciuglio

The authors study primitive-based inference for open-world compositional zero-shot learning, where all attribute–object pairings are candidates. They score images against separate attribute and object representations, using GDE directions or intrinsic manifold means instead of constructing pair representations. On MIT-States with CLIP ViT-L/14, GDE-Primitive reaches 10.90% HM and 11.98% UPA, exceeding GDE-Pair; UT-Zappos is an exception before whitening.

Summary by alphaXiv.

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