user12123/MA-Bench

MA-Bench是一个专为微动作理解设计的基准数据集，包含1,000个视频和12,000个结构化问答对。该数据集通过三层评估架构逐步考察微动作感知、关系理解和解释推理能力。此外，还提供了MA-Bench-Train，一个包含20.5K视频的大规模训练语料库，用于微调多模态大语言模型（MLLMs）。该工作旨在为MLLMs在理解细微微动作和人类相关行为方面建立基础基准。

MA-Bench is a benchmark designed for Micro-Action understanding, comprising 1,000 videos and 12,000 structured question–answer pairs. It features a three-tier evaluation architecture that progressively examines micro-action perception, relational comprehension, and interpretive reasoning. Additionally, it includes MA-Bench-Train, a large-scale training corpus with 20.5K videos for fine-tuning Multimodal Large Language Models (MLLMs). This work aims to establish a foundational benchmark for advancing MLLMs in understanding subtle micro-actions and human-related behaviors.