Research map / Vision-language and multimodal / Human and embodied modeling
Can You Read Me Now? Content Aware Rectification using Angle Supervision
Amir Markovitz, Inbal Lavi, Or Perel, Shai Mazor, Roee Litman
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 per-pixel angle loss (ablation) | 3D estimation model · Synthetic documents | edit distance: 0.324 -> 0.246 | Improved |
| restrict angle loss to text pixels with mask (ablation) | 3D estimation model · Synthetic documents | edit distance: 0.246 -> 0.244 | No clear effect |
| add angle confidence (magnitude weighting) (ablation) | 3D estimation model · Synthetic documents | edit distance: 0.244 -> 0.216 | Improved |
| add curvature estimation to angle loss (separate training) (ablation) | CREASE stage 1 · Synthetic documents | edit distance: 0.216 -> 0.187 | Improved |
| add curvature estimation to angle loss (E2E) (ablation) | CREASE · Synthetic documents | edit distance: 0.204 -> 0.178 | Improved |
| add angle loss on top of end-to-end training (ablation) | CREASE · Synthetic documents | edit distance: 0.223 -> 0.204 | Improved |
| add end-to-end fine-tuning with angles and curvature (ablation) | CREASE · Synthetic documents | edit distance: 0.187 -> 0.178 | Improved |
| add end-to-end fine-tuning alone (no angle or curvature) (ablation) | 3D estimator · Synthetic documents | edit distance: 0.324 -> 0.223 | Improved |
| add angle and curvature supervision (ablation) | CREASE · Synthetic documents | run-to-run std of edit distance: 0.169 -> 0.003 | Improved |
| add text segmentation auxiliary channel (ablation) | CREASE · Synthetic documents | edit distance: 0.178 -> 0.182 | No clear effect |
| replace DewarpNet with CREASE | document rectifier · Synthetic documents (Tesseract) | edit distance: 0.223 -> 0.178 | Improved |
| replace DewarpNet with CREASE | document rectifier · Synthetic documents (commercial OCR) | edit distance: 0.109 -> 0.103 | Improved |
Closest papers by meaning
- Foreground and Text-lines Aware Document Image Rectification (ICCV 2023 · similarity 0.88)
- Geometric Representation Learning for Document Image Rectification (ECCV 2022 · similarity 0.88)
- Multiview Rectification of Folded Documents (TPAMI 2017 · similarity 0.87)
- DocScanner: Robust Document Image Rectification with Progressive Learning (IJCV 2025 · similarity 0.87)
- ForCenNet: Foreground-Centric Network for Document Image Rectification (ICCV 2025 · similarity 0.85)
- ODM: A Text-Image Further Alignment Pre-training Approach for Scene Text Detection and Spotting (CVPR 2024 · similarity 0.83)
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.
