Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
RD-GAN: Few/Zero-Shot Chinese Character Style Transfer via Radical Decomposition and Rendering
Yaoxiong Huang, Mengchao He, Lianwen Jin, Yongpan Wang
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 |
|---|---|---|---|
| augment real training data with RD-GAN generated characters | character classifier · TKH recognition (D1) | accuracy (%): 82.45 -> 88.11 | Improved |
| use RD-GAN generated data instead of EMD generated data | character classifier · TKH recognition (D1) | accuracy (%): 87.21 -> 88.11 | Improved |
| use RD-GAN generated data instead of zi2zi generated data | character classifier · TKH recognition (D1) | accuracy (%): 85.63 -> 88.11 | Improved |
| add pix2pix generated data | character classifier · TKH recognition (D1) | accuracy (%): 82.45 -> 84.74 | Improved |
| add synthtext2014 generated data | character classifier · TKH recognition (D1) | accuracy (%): 82.45 -> 83.25 | Improved |
| train RD-GAN with 5 samples per class instead of 50 (D2 vs D1) | RD-GAN · TKH | SSIM: 0.9647 -> 0.9642 | No clear effect |
| train RD-GAN with 5 samples per class instead of 50 (D2 vs D1) | RD-GAN · TKH | L1 loss: 0.0088 -> 0.0089 | No clear effect |
| decompose characters into radicals to reach few-shot/zero-shot generalization | RD-GAN vs MC-GAN · TKH zero-shot (D3) | SSIM: 0.8633 -> 0.9112 | Improved |
| add lateral connections to the radical rendering module (ablation) | RD-GAN · TKH (Fig. 6 ablation) | L1 loss | Improved |
| add multi-level discriminator instead of single-level (ablation) | RD-GAN · TKH (Fig. 6 ablation) | SSIM | Improved |
| add additional losses (L2, total variation, content) (ablation) | RD-GAN · TKH (Fig. 6 ablation) | L1 loss / RMSE / SSIM | Improved |
Closest papers by meaning
- GAN-Based Unpaired Chinese Character Image Translation via Skeleton Transformation and Stroke Rendering (AAAI 2020 · similarity 0.89)
- Separating Style and Content for Generalized Style Transfer (CVPR 2018 · similarity 0.88)
- TET-GAN: Text Effects Transfer via Stylization and Destylization (AAAI 2019 · similarity 0.87)
- TE141K: Artistic Text Benchmark for Text Effect Transfer (TPAMI 2020 · similarity 0.87)
- A Multi-Style Chinese Characters Writing Intelligent Tool Based on Small-scale Training Data (AAAI 2025 · similarity 0.86)
- Puff-Net: Efficient Style Transfer with Pure Content and Style Feature Fusion Network (CVPR 2024 · similarity 0.84)
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.
