Reproduce “Moyun: A Diffusion-Based Model for Style-Specific Chinese Calligraphy Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kaiyuan Liu, Jiahao Mei, Hengyu Zhang, Yihuai Zhang, Daoguo Dong, Liang He
Moyun, a diffusion model, generates style-specific Chinese calligraphy characters by integrating a Vision Mamba architecture and a TripleLabel control mechanism. The model achieves fine-grained control over calligrapher, font, and character attributes, demonstrating high fidelity and zero-shot generalization to unseen stylistic combinations, supported by the new large-scale Mobao dataset.
Key insights
- Vision Mamba's State Space Model architecture effectively models complex contextual relationships and intricate stroke structures in calligraphy, outperforming traditional UNet-based approaches.
- Disentangling calligrapher, font, and character attributes into additive embeddings enables precise, computationally efficient multi-conditional control and robust zero-shot generalization.
- The novel SAM-based binarization technique significantly improves the quality of historical calligraphy images, making them suitable for training advanced deep generative models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.