Research map / Image generation / Image restoration
End-to-End Piece-Wise Unwarping of Document Images
Sagnik Das, Kunwar Yashraj Singh, Jon Wu, Erhan Bas, Vijay Mahadevan, Rahul Bhotika, Dimitris Samaras
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 the local piece-wise branch to the global branch in stitching (ablation) | piece-wise unwarping · DocUNet benchmark (130 images) | MS-SSIM: 0.4635 -> 0.4879 | Improved |
| add the global branch to the local branch in stitching (ablation) | piece-wise unwarping · DocUNet benchmark (130 images) | MS-SSIM: 0.4552 -> 0.4879 | Improved |
| add the local branch to the global branch (ablation) | piece-wise unwarping · DocUNet benchmark (130 images) | Local distortion (LD): 8.89 -> 9.23 | Hurt |
| add the global branch to the local branch (ablation) | piece-wise unwarping · DocUNet benchmark (130 images) | Local distortion (LD): 9.78 -> 9.23 | Improved |
| piece-wise unwarping versus global DewarpNet | document unwarping · DocUNet benchmark (130 images) | MS-SSIM: 0.4692 -> 0.4879 | Improved |
| piece-wise unwarping versus global DewarpNet | document unwarping · DocUNet benchmark (130 images) | Local distortion (LD): 8.98 -> 9.23 | Hurt |
| piece-wise unwarping versus global DewarpNet | document unwarping · DocUNet benchmark OCR (51 images) | CER: 0.3028 -> 0.3001 | Improved |
| piece-wise unwarping versus global DewarpNet | document unwarping · DocUNet benchmark OCR (51 images) | WER: 0.4368 -> 0.4302 | Improved |
| train SNet, PUNet and GSNet jointly end-to-end instead of freezing SNet and PUNet when training GSNet (ablation) | piece-wise unwarping · Doc3D validation | L1 error: ~2% L1 error increase in PUNet, ~1.6% L1 error decrease in GSNet | Hurt |
| add overlap between patches (ablation) | piece-wise unwarping · DocUNet benchmark | unwarping quality | No clear effect |
| use BatchNorm instead of Layer Normalization in PUNet (ablation) | PUNet · Doc3D training | generalization | Hurt |
| raise the reconstruction loss weights beta1 and gamma above 0.03 (ablation) | piece-wise unwarping · test-time unwarped images | image noise | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DocUNet — DocUNet: Document Image Unwarping via a Stacked U-Net (CVPR 2018) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DocUNet: Document Image Unwarping via a Stacked U-Net (CVPR 2018 · similarity 0.88)
- DewarpNet: Single-Image Document Unwarping With Stacked 3D and 2D Regression Networks (ICCV 2019 · similarity 0.85)
- Foreground and Text-lines Aware Document Image Rectification (ICCV 2023 · similarity 0.84)
- Axis-Aligned Document Dewarping (AAAI 2026 · similarity 0.83)
- Towards Real-World Document Parsing via Realistic Scene Synthesis and Document-Aware Training (CVPR 2026 · similarity 0.81)
- Predicting the Original Appearance of Damaged Historical Documents (AAAI 2025 · 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.
