Text-to-SVG Generation on SVGX-DwT-10k 1.0 (test)
100Validation AccuracyVectorFusion
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| VectorFusionTime(s)=680, #Token=100k, #Complex=2500, Zero-shot=false2025.05 | 100 | 25 | 30.1 | 5.7 | 18 | — | |
| DiffSketcherTime(s)=550, #Token=100k, #Complex=2500, Zero-shot=false2025.05 | 100 | 28.3 | 30.5 | 5.6 | 17.8 | — | |
| SVGDreamerTime(s)=1020, #Token=100k, #Complex=2500, Zero-shot=false2025.05 | 100 | 22.5 | 30.9 | 5.8 | 18.5 | — | |
| Reason-SVG (Full)Time(s)=12, #Token=∼3200, #Complex=145, Zero-shot=false2025.05 | 99.8 | 18.6 | 34.5 | 5.9 | 21.8 | 100 | |
| GPT-4oTime(s)=5, #Token=∼450, #Complex=85, Zero-shot=true2025.05 | 95.5 | 35.4 | 29.5 | 5.6 | 16.5 | — | |
| Claude 3.7 SonnetTime(s)=5, #Token=∼420, #Complex=80, Zero-shot=true2025.05 | 94.8 | 38.2 | 28.8 | 5.5 | 15.8 | — | |
| Gemini 2.5 ProTime(s)=16, #Token=∼400, #Complex=75, Zero-shot=true2025.05 | 94.5 | 40.6 | 28.1 | 5.4 | 15.65 | 100 | |
| o4-miniTime(s)=13, #Token=∼350, #Complex=65, Zero-shot=true2025.05 | 93 | 45.1 | 27 | 5.2 | 14.5 | 100 | |
| Qwen2.5-VL-72B-InstructTime(s)=4, #Token=∼400, #Complex=95, Zero-shot=false2025.05 | 92.8 | 34.3 | 29.2 | 5.4 | 16.3 | 91.8 | |
| DeepSeek-R1Time(s)=21, #Token=∼380, #Complex=90, Zero-shot=false2025.05 | 92.5 | 32.5 | 29 | 5.3 | 16.2 | 100 | |
| SFT-DwT (w/o RL)Time(s)=8, #Token=∼1500, #Complex=120, Zero-shot=false2025.05 | 89 | 21.2 | 31 | 5.7 | 19.5 | 92.3 | |
| LLM4SVG (Qwen2.5-7B-Instruct)Time(s)=25, #Token=∼215, #Complex=45, Zero-shot=false2025.05 | 76 | 30.7 | 29.3 | 5.2 | 16.8 | — | |
| SFT-vanillaTime(s)=5, #Token=∼300, #Complex=55, Zero-shot=false2025.05 | 75 | 28.1 | 28.5 | 5.3 | 17.5 | — | |
| StarVector (SD sampling + Img2SVG)Time(s)=90, #Token=∼370, #Complex=70, Zero-shot=false2025.05 | 72 | 35.8 | 28.8 | 5.1 | 16 | — |