VGQA
收藏arXiv2025-09-30 收录
下载链接：
https://vgbench.github.io
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个旨在评估模型在矢量图形理解能力方面的基准测试，它包含了一系列针对不同矢量图形类别的定制任务。数据集涵盖了SVG、TikZ和Graphviz等样本，并使用多项选择题和图像生成指标（如FID和CLIP得分）进行评估。该数据集的规模包括4279个理解样本和5845个生成样本，任务涵盖了视觉理解和视觉生成。
This dataset is a benchmark developed to evaluate models' capabilities in vector graphics understanding. It includes a series of customized tasks tailored for different vector graphic categories. The dataset covers samples in formats such as SVG, TikZ and Graphviz, and uses multiple-choice questions as well as image generation metrics (including FID and CLIP scores) for evaluation. It consists of 4279 understanding samples and 5845 generation samples, with tasks covering both visual understanding and visual generation.
提供机构：
OpenAI and human annotators
