Research map / Detection and segmentation / CNN features
End-to-End Handwritten Paragraph Text Recognition Using a Vertical Attention Network
Denis Coquenet, Clément Chatelain, Thierry Paquet
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 line-level pretraining of the encoder (ablation) | VAN · IAM paragraph | CER (%): 7.06 -> 4.45 | Improved |
| add line-level pretraining of the encoder (also faster training) (ablation) | VAN · IAM paragraph | CER (%): 8.06 -> 4.45 | Improved |
| add dropout when training from scratch (ablation) | VAN · IAM paragraph | CER (%): 8.06 -> 7.06 | Improved |
| use cross-dataset pretraining instead of line-level pretraining | VAN · RIMES paragraph | CER (%): 1.91 -> 1.97 | No clear effect |
| use cross-dataset pretraining instead of line-level pretraining | VAN · IAM paragraph | CER (%): 4.45 -> 4.55 | No clear effect |
| use cross-dataset pretraining instead of line-level pretraining | VAN · READ 2016 paragraph | CER (%): 3.59 -> 4.08 | Hurt |
| add learned stop instead of fixed stop (ablation) | VAN · IAM paragraph | CER (%): 4.41 -> 4.45 | No clear effect |
| add learned stop instead of fixed stop (train time) (ablation) | VAN · IAM paragraph | training time: 1.20 d -> 0.63 d | Improved |
| process whole paragraphs with VAN instead of two-step segmentation then OCR | VAN · IAM paragraph | CER (%): 6.55 -> 4.45 | Improved |
| process whole paragraphs instead of isolated lines | VAN · IAM | CER (%): 4.97 -> 4.54 | Improved |
| process whole paragraphs instead of isolated lines | VAN · RIMES | CER (%): 3.08 -> 2.15 | Improved |
| add Mix Dropout instead of standard dropout (ablation) | line-level model · IAM line-level | CER (%): 5.24 -> 5.01 | Improved |
Closest papers by meaning
- Joint Line Segmentation and Transcription for End-to-End Handwritten Paragraph Recognition (NeurIPS 2016 · similarity 0.89)
- DAN: A Segmentation-Free Document Attention Network for Handwritten Document Recognition (TPAMI 2023 · similarity 0.89)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.88)
- Decoupled Attention Network for Text Recognition (AAAI 2020 · similarity 0.87)
- OrigamiNet: Weakly-Supervised, Segmentation-Free, One-Step, Full Page Text Recognition by learning to unfold (CVPR 2020 · similarity 0.87)
- General Detection-based Text Line Recognition (NeurIPS 2024 · similarity 0.87)
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.
