Research map / Vision-language and multimodal / Human and embodied modeling
D2Dewarp: Dual Dimensions Geometric Representation Learning Based Document Image Dewarping
Heng Li, Xiangping Wu, Qingcai Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add HV Fusion Module (ablation) | D2Dewarp · DocUNet (60 images) | CER: 0.1439 -> 0.1338 | Improved |
| add HV Fusion Module (ablation) | D2Dewarp · DocUNet | ED (60 images): 401.1 -> 351.1 | Improved |
| add HV Fusion Module (ablation) | D2Dewarp · DIR300 | ED: 509.8 -> 466.9 | Improved |
| add HV Fusion Module (ablation) | D2Dewarp · DocReal | MS-SSIM: 0.56 -> 0.58 | Improved |
| add HV Fusion Module (ablation) | D2Dewarp · DocUNet | LD: 7.6 -> 7.7 | Hurt |
| add HV Fusion Module (ablation) | D2Dewarp · DocReal | CER: 0.2665 -> 0.2588 | Improved |
| add V-Line to H-Line only (ablation) | D2Dewarp · DocReal | LD/AD: 9.01/0.253 -> 8.69/0.227 | Improved |
| add H-Line to V-Line only (ablation) | D2Dewarp · DocReal | ED: 201.12 -> 191.25 | Improved |
| train DocScanner on DocDewarpHV instead of Doc3D | D2Dewarp · DocUNet | MS-SSIM: 0.52 -> 0.47 | Hurt |
| D2Dewarp vs DocReal (best prior in DocReal benchmark) | D2Dewarp · DocReal | MS-SSIM: 0.56 -> 0.58 | Improved |
| D2Dewarp vs DocReal on OCR metrics | D2Dewarp · DocReal | ED: 184.54 -> 191.25 | Hurt |
| D2Dewarp vs RDGR, DocGeoNet, FTDR | D2Dewarp · DocUNet | CER: at least 9.5% and 11.3% on 50 and 60 images | Improved |
Closest papers by meaning
- Axis-Aligned Document Dewarping (AAAI 2026 · similarity 0.90)
- DewarpNet: Single-Image Document Unwarping With Stacked 3D and 2D Regression Networks (ICCV 2019 · similarity 0.88)
- Revisiting Document Image Dewarping by Grid Regularization (CVPR 2022 · similarity 0.86)
- Geometric Representation Learning for Document Image Rectification (ECCV 2022 · similarity 0.85)
- DocFormerv2: Local Features for Document Understanding (AAAI 2024 · similarity 0.83)
- DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks (CVPR 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
