Reproduce “VL-Dedup: Graded Multimodal Entity Resolution for Image-Text Data Deduplication” 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
Zhenxi Liu, Guanze Chen, Dehui Yu, Xingyu Wang, Bingxu Hou, Chengwei Li, Tao Ma, Ke Wang and 1 more
VL-Dedup curates image–text corpora by retrieving candidate pairs, classifying links, and forming clusters subject to a size cap. It uses vision-language judgments and CLIPScore to flag mismatches and compare captions, retaining complementary details or holding uncertain records for review. On the original test records, a cap of five raised HGB’s all-pairs F1 from 0.0023 to 0.9779; an audit found fewer lost facts than keeping one record per cluster.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.