Research map / Detection and segmentation / Weakly supervised detection and segmentation
OrigamiNet: Weakly-Supervised, Segmentation-Free, One-Step, Full Page Text Recognition by learning to unfold
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use OrigamiNet to unfold full-page input into a CTC-compatible sequence without localization labels | GTR-12 · ICDAR2017 HTR validation | CER: 8.18 -> 6.80 | Helped |
| Use OrigamiNet with a larger input scale | GTR-8 OrigamiNet · IAM paragraph test | CER (%): 5.6 -> 5.5 | Helped |
| Use GTR-12 OrigamiNet instead of GTR-8 OrigamiNet | OrigamiNet · IAM paragraph test | CER (%): 5.5 -> 4.7 | Helped |
| Add end-to-end layer normalization (ablation) | OrigamiNet with VGG, ResNet-26, ResNet-66, ResNet-74, and GTR-8 · IAM test | CER: 51.37 -> 34.55; 10.03 -> 7.238; 8.925 -> 6.373; 76.9 -> 6.128; 72.4 -> 5.639 | Helped |
| Increase final sequence length L2 from 700 to 1100 (ablation) | ResNet-26 OrigamiNet, L1=450 · IAM test | CER: 8.121 -> 7.238 | Helped |
| Increase final sequence length L2 from 1100 to 1500 (ablation) | ResNet-26 OrigamiNet, L1=450 · IAM test | CER: 7.238 -> 7.449 | Hurt |
| Increase final sequence length L2 from 700 to 1500 (ablation) | VGG OrigamiNet, L1=450 · IAM test | CER: 43.14 -> 30.34 | Helped |
| Use final width 31 instead of 62 (ablation) | ResNet-26 OrigamiNet, L1=450, L2=1100 · IAM test | CER: 19.9 -> 9.128 | Helped |
| Vary final width across 62, 31, 15, 8, and 3 (ablation) | VGG OrigamiNet, L1=450, L2=1100 · IAM test | CER: 25.98; 17.41; 37.4; 34.55; 24.21 | Mixed |
| Use final width 62 instead of 8 (ablation) | ResNet-26 OrigamiNet, L1=450, L2=1100 · IAM test | CER: 7.238 -> 19.9 | Hurt |
| Train on heavily touching text lines produced by reducing interline spacing | GTR-12 OrigamiNet · IAM compact-line variant | CER: 6.5% | Mixed |
| Train on rotated and warped text lines | GTR-12 OrigamiNet · IAM rotated-and-warped variant | CER: 6.2% | Mixed |
Closest papers by meaning
- General Detection-based Text Line Recognition (NeurIPS 2024 · similarity 0.86)
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning (NeurIPS 2025 · similarity 0.86)
- MetaWriter: Personalized Handwritten Text Recognition Using Meta-Learned Prompt Tuning (CVPR 2025 · similarity 0.86)
- SRFormer: Text Detection Transformer with Incorporated Segmentation and Regression (AAAI 2024 · similarity 0.86)
- Bridging the Gap Between End-to-End and Two-Step Text Spotting (CVPR 2024 · similarity 0.85)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.85)
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.