Research map / Image generation / Text-to-image diffusion
ForCenNet: Foreground-Centric Network for Document Image Rectification
Peng Cai, Qiang Li, Kaicheng Yang, Dong Guo, Jia Li, Nan Zhou, Xiang An, Ninghua Yang, Jiankang Deng
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 mask-guided transformer decoder (ablation) | ForCenNet · DocUNet (ablation) | MS-SSIM: 0.558 -> 0.571 | Improved |
| add mask-guided transformer decoder (ablation) | ForCenNet · DocUNet (ablation) | LD: 5.44 -> 4.95 | Improved |
| add curvature consistency loss (ablation) | ForCenNet · DocUNet (ablation) | MS-SSIM: 0.565 -> 0.571 | Improved |
| add curvature consistency loss (ablation) | ForCenNet · DocUNet (ablation) | CER: 0.169 -> 0.141 | Improved |
| add both mask-guided decoder and curvature loss (ablation) | ForCenNet · DocUNet (ablation) | LD: 7.06 -> 4.95 | Improved |
| make foreground segmentation differentiable instead of frozen separately trained model (ablation) | ForCenNet · DocUNet (ablation) | MS-SSIM: 0.468 -> 0.571 | Improved |
| make foreground segmentation differentiable instead of frozen separately trained model (ablation) | ForCenNet · DocUNet (ablation) | CER: 0.212 -> 0.141 | Improved |
| scale distortion samples per image from x1 to x100 (ablation) | ForCenNet · DocUNet (65 images) | MS-SSIM: 0.449 -> 0.530 | Improved |
| scale distortion samples from x100 to x1000 (ablation) | ForCenNet · DocUNet (65 images) | MS-SSIM: 0.530 -> 0.571 | Improved |
| scale distortion samples from x1000 to x5000 (ablation) | ForCenNet · DocUNet (65 images) | MS-SSIM: 0.571 -> 0.569 | No clear effect |
| coarser line sampling interval 4 to 32 pixels (ablation) | ForCenNet · DocUNet | MS-SSIM: 0.582 -> 0.569 | Hurt |
| train on Doc3D undistorted images instead of DocUNet/DIR300 images (ablation) | ForCenNet · DocUNet | MS-SSIM: 0.582 -> 0.579 | Mixed |
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.
- DocRes — DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks (CVPR 2024) · better on 2 of 2 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
- Foreground and Text-lines Aware Document Image Rectification (ICCV 2023 · similarity 0.87)
- Geometric Representation Learning for Document Image Rectification (ECCV 2022 · similarity 0.87)
- DocUNet: Document Image Unwarping via a Stacked U-Net (CVPR 2018 · similarity 0.85)
- Can You Read Me Now? Content Aware Rectification using Angle Supervision (ECCV 2020 · similarity 0.85)
- DocScanner: Robust Document Image Rectification with Progressive Learning (IJCV 2025 · similarity 0.83)
- In-Context Matting (CVPR 2024 · similarity 0.80)
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.
