A-Bench
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/Q-Future/A-Bench
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为A-Bench，旨在评估大型多模态模型（LMMs）在评估人工智能生成图像（AIGIs）方面的有效性。它包含了来自16个文本到图像模型生成的2,864张AIGIs，每张图像都配有人类专家标注的问题和答案。该基准测试结构旨在评估高级语义理解能力和低级视觉质量感知能力，评估范围覆盖了18个领先的多模态模型。该数据集规模为2,864张人工智能生成的图像，其任务是评估人工智能生成图像。
This dataset, named A-Bench, is designed to evaluate the effectiveness of Large Multimodal Models (LMMs) in assessing AI-Generated Images (AIGIs). It contains 2,864 AIGIs generated by 16 text-to-image models, with each image paired with human-expert annotated questions and answers. The benchmark framework aims to evaluate both high-level semantic understanding capabilities and low-level visual quality perception capabilities, and covers 18 state-of-the-art multimodal models. With a total of 2,864 AI-generated images, the core task of this dataset is to evaluate AI-generated images.
