Reproduce “Telligram: Text-Driven Calligram Generation via Diffusion-Guided Skeleton Optimization” 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
Tianci Shi, Pengfei Xu
Telligram generates calligrams from a word and semantic prompt without an input image or contour. Its two-stage pipeline uses diffusion guidance to form an occupancy shape, then reconstructs letters within it using geometric fitting and OCR-guided selection. In a comparison, hierarchical updating reaches a CLIPScore of 34.314, slightly above alternatives, but has higher area-imbalance indicators and does not establish readability.
Key insights
- Decoupling semantic shape formation from readability-constrained glyph realization into distinct stages effectively mitigates the conflict between these two goals, enhancing optimization stability.
- Hierarchical gradient projection, which prioritizes coarse affine deformations before applying graph-smooth local skeleton updates, is crucial for improving semantic alignment and overall stability.
- Leveraging pretrained diffusion models for robust semantic occupancy prediction, complemented by geometric post-processing, enables a training-free approach to text-only calligram generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.