Reproduce “J-CRe3: A Japanese Conversation Dataset for Real-world Reference Resolution” 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
Nobuhiro Ueda, Hideko Habe, Yoko Matsui, Akishige Yuguchi, Seiya Kawano, Yasutomo Kawanishi, Sadao Kurohashi, Koichiro Yoshino
J-CRe3 introduces a Japanese dialogue-and-egocentric-video dataset and grounding task that links utterance phrases and semantic roles to scene objects, including omitted and indirect references. It contains 93 scenario-based videos and 2,131 utterances. After adaptation, MDETR achieved Recall@1 of 0.410 for direct phrase grounding, versus 0.005 without adaptation; this does not resolve the full task or demonstrate robot action.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.