Tag
DataVista 是首个针对数据视频(datavideo)理解的基准测试,包含 961 个真实数据视频和 6,775 道评估题,采用三级能力框架(数据感知、时序推理、叙事理解)。对 19 个主流多模态大模型的评测显示,最佳模型 Gemini-3.1-Pro 仅有 70.0% 准确率,远低于人类专家水平。
This paper investigates how vision-language models read exact values from vertical bar charts using controlled counterfactual activation patching, revealing insights into internal computations and differences between models like Qwen and InternVL.