Reproduce “Rethinking Supervision in Document Unwarping: A Self-consistent Flow-free Approach” 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
Shaokai Liu, Hao Feng, Wengang Zhou
DRNet rectifies photographed documents by predicting a dense warp and training against undistorted images with pixel-level and VGG-feature losses, rather than ground-truth warp maps. It also encourages consistency between predictions for flipped inputs. On the DocUNet Benchmark, it reports a character error rate of 0.1649, versus 0.1717 for RDGR, on 50 text-heavy images; incomplete or absent page boundaries remain challenging.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.