Reproduce “FrameLite: Decision-Oriented Aesthetic Cropping with a Compact Vision-Language Model” 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
Jianze Li, Kai Liu, Ziqing Zhang, Weihang Zhang, Yong Guo, Linghe Kong, Yulun Zhang
FrameLite trains a compact image-cropping model to return one coordinate crop without candidate search at inference. It learns localization and framing, imitates high-quality crops selected from its own samples, then refines samples by comparing their quality with its current greedy crop. On GAICv2, it scores 91.4% versus Venus’s 83.8% at IoU 0.85, a benchmark comparison, not a speed measurement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.