Reproduce “DREAM: Document Reconstruction via End-to-end Autoregressive 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
Xin Li, Mingming Gong, Yunfei Wu, Jianxin Dai, Antai Guo, Xinghua Jiang, Haoyu Cao, Yinsong Liu and 2 more
DREAM reconstructs document-page images as ordered records combining each element’s category, location, and transcribed content. It uses a Swin image encoder and a parallel decoder that generates element sequences, with autoregressive decoding within each element. On DocRec1K, DREAM scores 91.4 DSM and 92.7 NED, versus 85.2 and 88.6 for adapted Nougat-base; this result applies to the defined 1,000-image benchmark.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.