Reproduce “UniMore: Fine-Grained Supervision for Unified Understanding and Generation” 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
Haoyue Bai, Yinyi Luo, Yicheng He, Qingsong Wen, Sharon Li, Jindong Wang
UniMore trains a model to preserve visual details in image generation and captioning. It uses image-caption pairs to match descriptive phrases with grounded crops, and ranks full captions above versions omitting or altering details. On BAGEL, detail retention rises from 28.22% to 29.77%, while joint object-and-color correctness rises from 53.75% to 62.00%. These gains are reported on specific diagnostics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.