Reproduce “U-VLM: Hierarchical Vision Language Modeling for Report 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
Pengcheng Shi, Minghui Zhang, Kehan Song, Jiaqi Liu, Yun Gu, Xinglin Zhang
U-VLM generates 3D CT reports with a shared U-Net encoder trained on segmentation, classification, and reports, injecting features from different encoder depths into language layers. On CT-RATE, it reaches macro-F1 0.414 on 18 abnormality labels extracted from reports by RadBERT at a 0.5 threshold, and BLEU-mean 0.349. These benchmark results do not establish clinical correctness; examples include omissions and altered findings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.