Researchers have developed VFIG, a new vision-language model (VLM) designed to convert rasterized images into Scalable Vector Graphics (SVG) format. This advancement addresses the common issue of lost original vector files, which makes technical illustrations difficult to edit. VFIG is trained on VFIG-Data, the largest dataset of its kind, and evaluated using VFIG-Bench, a new benchmark that assesses structural correctness beyond simple visual similarity. The model demonstrates state-of-the-art open-source performance, outperforming existing baselines and nearing the capabilities of proprietary models like Claude Sonnet 4.6 and GPT-5.2. AI
IMPACT This research could significantly improve the workflow for technical illustration and diagram creation by automating the conversion of raster images to editable vector formats.
RANK_REASON The item is an academic paper detailing a new model and dataset for a specific technical task. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Claude Sonnet 4.6
- Gemini 3
- Gemini-Judge
- GPT-5.2
- SVG
- VFIG
- VFIG-Bench
- VFIG-Bench-OOD
- VFIG-Data
- vision-language model
- VLM-Judge
- Xunmei Liu
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
