SVG Generation Benchmark (Static)

Summary

The SVG Generation Benchmark is a ~189k-row human preference dataset for evaluating text-to-SVG (vector graphics) generation models. Each row contains a text prompt, two SVG outputs from different models (e.g., Claude, Gemini), rendered images, and weighted human preference votes across multiple criteria (preference, coherence, alignment). It enables comparison and ranking of SVG generation models based on real human judgments collected via the Rapidata platform.

Keywords

svg vector-graphics text-to-svg benchmark human-preference evaluation rapidata image-generation

Topics

Vision / Text-to-Image Evaluation

Research notes

• Built by Rapidata (rapidata.ai). Includes demographic metadata for each vote (country, language, age, gender). Compares models like Claude 3.5 Sonnet and Claude 3.5 Thinking.