MMSVGBench Multimodal Vector Graphics Generation Benchmark Dataset
MMSVG-Bench is a comprehensive benchmark designed for multimodal SVG generation tasks, released in 2025 by Fudan University in collaboration with StepFun. The related paper is titled "".OmniSVG: A Unified Scalable Vector Graphics Generation ModelThe dataset has been selected for NeurIPS 2025 Datasets and Benchmarks, aiming to fill the gap in the current field of vector graphics generation, which lacks a unified, open, and standardized test set.
The dataset contains 600 test samples, divided into two subsets according to task type:
- image2svg (300 entries): The input is an image, and the goal is to generate the corresponding SVG.
- text2svg (300 entries): The input is a text description, and the model needs to generate SVG based on the text.
Each sample contains complete input information and metadata, including a unique identifier, image or text input, task type, category label (such as icon or illustration), and the URL of the data source. Specifically, the image field of the text2svg sample is empty, and the text field of the image2svg sample is empty.
Build AI with AI
From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.