Research map / Detection and segmentation / CNN features
Learning Spatial-Semantic Context with Fully Convolutional Recurrent Network for Online Handwritten Chinese Text Recognition
Zecheng Xie, Zenghui Sun, Lianwen Jin, Hao Ni, Terry Lyons
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 |
|---|---|---|---|
| raise path signature truncation from Sig0 to Sig3 (ablation) | FCRN · Dataset-ICDAR | correct rate (CR, %): 90.18 -> 92.35 | Improved |
| raise path signature truncation from Sig0 to Sig3 (ablation) | FCRN · Dataset-CASIA | correct rate (CR, %): 92.59 -> 94.52 | Improved |
| raise path signature from Sig2 to Sig3 (ablation) | FCRN · Dataset-ICDAR | correct rate (CR, %): 92.25 -> 92.35 | Improved |
| add a second receptive-field scale (2C-FCRN) (ablation) | FCRN · Dataset-ICDAR | correct rate (CR, %): 92.25 -> 93.08 | Improved |
| add a third receptive-field scale (3C-FCRN) (ablation) | FCRN · Dataset-ICDAR | correct rate (CR, %): 92.25 -> 93.53 | Improved |
| add three receptive-field scales (3C-FCRN) (ablation) | FCRN · Dataset-CASIA | correct rate (CR, %): 94.37 -> 95.16 | Improved |
| use same-scale receptive fields with similar size (FCRN-2) (ablation) | FCRN · Dataset-ICDAR | correct rate (CR, %): 92.25 -> 92.40 | No clear effect |
| use same-scale receptive fields with similar size (FCRN-3) (ablation) | FCRN · Dataset-CASIA | correct rate (CR, %): 94.37 -> 94.17 | No clear effect |
| add bidirectional rather than unidirectional LSTM in implicit LM (SLD) (ablation) | 3C-FCRN · Dataset-ICDAR | correct rate (CR, %): 94.59 -> 95.19 | Improved |
| add implicit language model (B_SLD) | 3C-FCRN · Dataset-ICDAR | correct rate (CR, %): 93.53 -> 95.19 | Improved |
| add trigram LM on top of implicit LM (ablation) | 3C-FCRN+B_SLD · Dataset-ICDAR | correct rate (CR, %): 95.19 -> 97.15 | Improved |
| use larger SLD corpus rather than PFR for implicit LM (ablation) | 3C-FCRN+B · Dataset-ICDAR | correct rate (CR, %): 94.90 -> 95.19 | Improved |
Closest papers by meaning
- Drawing and Recognizing Chinese Characters with Recurrent Neural Network (TPAMI 2017 · similarity 0.86)
- PageNet: Towards End-to-End Weakly Supervised Page-Level Handwritten Chinese Text Recognition (IJCV 2022 · similarity 0.85)
- Joint Line Segmentation and Transcription for End-to-End Handwritten Paragraph Recognition (NeurIPS 2016 · similarity 0.84)
- Chinese Street View Text: Large-Scale Chinese Text Reading With Partially Supervised Learning (ICCV 2019 · similarity 0.83)
- Recursive Recurrent Nets With Attention Modeling for OCR in the Wild (CVPR 2016 · similarity 0.83)
- Fully Convolutional Networks for Continuous Sign Language Recognition (ECCV 2020 · similarity 0.82)
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.
