Research map / Probabilistic and scientific modeling / Variational and generative sampling
A Scalable Vector Graphics Latent Space
Leonardo Zini, Elia Frigieri, Lorenzo Baraldi
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 |
|---|---|---|---|
| use BPE tokenizer instead of command-based tokenizers (ablation) | SLS · path reconstruction | MSE-sim: 75.60 / 79.26 -> 90.83 | Improved |
| model whole paths instead of fixed-length token blocks (ablation) | SLS · path reconstruction | mIoU: 64.31 -> 78.92 | Improved |
| use EOS pooling instead of CLS token (ablation) | SLS · path reconstruction | MSE-sim: 90.16 -> 90.83 | Improved |
| add KL loss to CLS pooling (ablation) | SLS · path reconstruction | MSE-sim: 90.16 -> 90.17 | No clear effect |
| add l2 norm regularization or learnable logit scale (ablation) | SLS · path reconstruction | MSE-sim: 90.83 -> 86.51 (l2); 71.98 (logit) | Hurt |
| replace XML tokens with SLS path embeddings for captioning | LLM captioner · SVG captioning | BLEU5: 12.31 -> 46.89 (Llama 3.2 1B) | Improved |
| replace XML tokens with SLS embeddings | LLM captioner · captioning cost | context length: 2432.02 -> 15.56 tokens | Improved |
| use SLS instead of DeepSVG as encoder | LLM captioner · SVG captioning | BLEU5: 18.77 -> 46.89 (Llama 3.2 1B) | Improved |
| use SLS instead of raster encoders for retrieval | path retrieval · path retrieval | MSE-Sim: 98.86 -> 89.95 | Hurt |
| use SLS instead of raster encoders for retrieval | path retrieval · path retrieval | BLEU5: 10.77 -> 20.91 | Improved |
| evaluate out-of-domain UniSVG | SLS · UniSVG | MSE-sim: 90.83 -> 86.67 | Hurt |
Closest papers by meaning
- Chat2SVG: Vector Graphics Generation with Large Language Models and Image Diffusion Models (CVPR 2025 · similarity 0.88)
- SVGBuilder: Component-Based Colored SVG Generation with Text-Guided Autoregressive Transformers (AAAI 2025 · similarity 0.87)
- VFIG: Vectorizing Complex Figures in SVG with Vision-Language Models (NeurIPS 2026 · similarity 0.87)
- StarVector: Generating Scalable Vector Graphics Code from Images and Text (AAAI 2025 · similarity 0.87)
- OmniSVG: A Unified Scalable Vector Graphics Generation Model (NeurIPS 2025 · similarity 0.87)
- InternSVG: Towards Unified SVG Tasks with Multimodal Large Language Models (ICLR 2026 · similarity 0.86)
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.
