Every paper that un-crinkles a photographed page, in one table

What it does A curated list of research papers on document image rectification — the task of taking a warped, curled photo of a page and producing a flat, readable scan. It covers roughly 45 papers from 2018 through 2026, organized by year and venue, plus a separate table of datasets and a demos section. Each entry links the paper, code repo where one exists, and rectified results on the DocUNet Benchmark Dataset.

The interesting bit Read the table top to bottom and you can watch the field’s fashions change: stacked U-Nets and 3D regression in 2018–2019, control points and grid regularization around 2021–2022, then transformers, generalist restoration models, and finally coordinate-based diffusion in 2025. Papers marked with an asterisk also handle illumination correction, so you can filter for the shadow-and-glare side of the problem.

Key highlights

• Spans 2018 (DocUNet) to 2026 entries like BookNet, which rectifies book spreads via cross-page attention.
• Code links for many entries — DocTr, DocGeoNet, DocTr-Plus, DocReal, UVDoc, DocRes, and more.
• Separate datasets table distinguishing real vs. synthetic benchmarks (DocUNet, DewarpNet, Doc3D, Inv3D, etc.).
• Result links point to rectified outputs on the shared DocUNet benchmark, easing cross-paper comparison.
• Maintained by the author of DocTr/DocGeoNet/DocScanner, so the lineage is well covered.

Caveats

• It’s a list, not a survey — no commentary on which methods actually work best.
• Some entries are arXiv-only or link to Baidu pan downloads, which may be less accessible.

Verdict Essential bookmark if you work on document scanning, OCR preprocessing, or camera-capture enhancement. If you just want a working dewarper, skip straight to the repos with code and results links.