Research map / Architectures and efficiency / RNNs and normalization
Drawing and Recognizing Chinese Characters with Recurrent Neural Network
Xu-Yao Zhang, Fei Yin, Yan‐Ming Zhang, Cheng‐Lin Liu, Yoshua Bengio
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 |
|---|---|---|---|
| replace LSTM units with GRU units (1 layer) | bidirectional RNN · ICDAR-2013 | test accuracy %: 97.67 -> 97.71 | No clear effect |
| replace LSTM with GRU units (2 layers) | bidirectional RNN · ICDAR-2013 | test accuracy %: 97.70 -> 97.76 | No clear effect |
| replace LSTM with GRU units | bidirectional RNN · ICDAR-2013 | test time per character (ms): 0.3792 -> 0.2949 | Improved |
| replace LSTM with GRU units (3 layers) | bidirectional RNN · ICDAR-2013 | test accuracy %: 97.80 -> 97.77 | No clear effect |
| replace LSTM with GRU units | bidirectional RNN · ICDAR-2013 | memory: 11.00MB -> 9.06MB | Improved |
| stack two bidirectional layers instead of one (GRU) | bidirectional GRU · ICDAR-2013 | test accuracy %: 97.71 -> 97.76 | Improved |
| stack three bidirectional layers instead of one (LSTM) | bidirectional LSTM · ICDAR-2013 | test accuracy %: 97.67 -> 97.80 | Improved |
| stack three layers instead of one (LSTM) | bidirectional LSTM · ICDAR-2013 | training time (h): 95.16 -> 216.75 | Hurt |
| predict from a single sub-sequence from sequential dropout instead of full sequence (ablation) | NET4 · ICDAR-2013 | test accuracy %: 97.76 -> 96.54 | Hurt |
| ensemble 30 sub-sequences from input dropout at test time (ablation) | NET4 · ICDAR-2013 | test accuracy %: 97.76 -> 97.89 | Improved |
| ensemble 30 sub-sequences at test time (ablation) | NET5 · ICDAR-2013 | test accuracy %: 97.80 -> 97.96 | Improved |
| average six separately trained networks | NET1 to NET6 · ICDAR-2013 | test accuracy %: 97.76 -> 98.15 | Improved |
Closest papers by meaning
- Learning Spatial-Semantic Context with Fully Convolutional Recurrent Network for Online Handwritten Chinese Text Recognition (TPAMI 2017 · similarity 0.86)
- PageNet: Towards End-to-End Weakly Supervised Page-Level Handwritten Chinese Text Recognition (IJCV 2022 · similarity 0.85)
- Stroke Extraction of Chinese Character Based on Deep Structure Deformable Image Registration (AAAI 2023 · similarity 0.84)
- A Neural Representation of Sketch Drawings (ICLR 2018 · similarity 0.84)
- Fully Convolutional Network Based Skeletonization for Handwritten Chinese Characters (AAAI 2018 · similarity 0.84)
- A Multi-Style Chinese Characters Writing Intelligent Tool Based on Small-scale Training Data (AAAI 2025 · 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.
