Research map / Vision-language and multimodal / Human and embodied modeling
DuetSVG: Unified Multimodal SVG Generation with Internal Visual Guidance
Peiying Zhang, Nanxuan Zhao, Matthew Fisher, Yiran Xu, Jing Liao, Difan Liu
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 |
|---|---|---|---|
| add internal visual guidance (image tokens before SVG tokens) (ablation) | DuetSVG · SVG-Hub-5M test (text-to-SVG) | FID: 51.482 -> 33.574 | Improved |
| add internal visual guidance (ablation) | DuetSVG · SVG-Hub-5M test (image-to-SVG) | DINO: 0.939 -> 0.962 | Improved |
| add text-to-image pretraining stage (ablation) | DuetSVG · SVG-Hub-5M test (text-to-SVG) | FID: 36.953 -> 33.574 | Improved |
| add test-time scaling with image-guided SVG resampling (ablation) | DuetSVG · SVG-Hub-5M test (text-to-SVG) | FID: 35.066 -> 33.574 | Improved |
| add test-time scaling with image-guided SVG resampling (ablation) | DuetSVG · SVG-Hub-5M test (image-to-SVG) | LPIPS: 0.082 -> 0.075 | Improved |
| use image-guided SVG resampling instead of best-of-N (ablation) | DuetSVG · SVG-Hub-5M test (text-to-SVG) | sampling cost at given CLIP score | Improved |
| fine-tune open-source VLM baselines on SVG-Hub-5M | OmniSVG-3B · SVG-Hub-5M test (text-to-SVG) | FID: 93.172 -> 58.237 | Improved |
| use DuetSVG with multimodal output vs fine-tuned Qwen3-VL-8B text-only | SVG generation backbone · SVG-Hub-5M test (text-to-SVG) | FID: 43.720 -> 35.066 | Improved |
| use DuetSVG instead of VTracer for image-to-SVG reconstruction | image-to-SVG · SVG-Hub-5M test (image-to-SVG) | DINO / LPIPS: DINO 0.968 -> 0.962; LPIPS 0.071 -> 0.075 | Mixed |
Closest papers by meaning
- InternSVG: Towards Unified SVG Tasks with Multimodal Large Language Models (ICLR 2026 · similarity 0.91)
- Chat2SVG: Vector Graphics Generation with Large Language Models and Image Diffusion Models (CVPR 2025 · similarity 0.90)
- OmniSVG: A Unified Scalable Vector Graphics Generation Model (NeurIPS 2025 · similarity 0.89)
- StarVector: Generating Scalable Vector Graphics Code from Images and Text (AAAI 2025 · similarity 0.88)
- VFIG: Vectorizing Complex Figures in SVG with Vision-Language Models (NeurIPS 2026 · similarity 0.87)
- Empowering LLMs to Understand and Generate Complex Vector Graphics (CVPR 2025 · similarity 0.87)
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.
