Reproduce “ZoomCap: Evidence-Routed Visual Zooming for Recoverable Visual Information Transfer” 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
Hao Shi, Jiafeng Wang, Yuzhuo Fu, Tao Zhang, Jian Bai, Juelu Zhang, Qing Cui, Chao Huang
ZoomCap improves image captions by drafting a global description, then selectively gathering missing visual evidence through targeted zooming, OCR, and conditional knowledge search, while checking claims before fusion. On FactCycle-250, it achieves 52.4 transferred claims per image versus 31.5 for ScaleCap. This reconstruction-based measure uses one fixed image generator and excludes text-heavy images, so it does not test OCR.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.