ActionArt

The dataset is introduced in our paper titled "ActionArt: Advancing Multimodal Large Models for Fine-Grained Human-Centric Video Understanding" (https://arxiv.org/abs/2504.18152). It comprises a training set and a test set. The training set includes video names along with manually annotated video captions. The test set is further divided into multiple subsets, each documented in a JSON file that contains the video names and their corresponding manually annotated QA. This dataset utilizes videos from the Movid dataset, and we encourage you to refer to their work to access the original videos. 可以通过如下GIT Clone命令，或者ModelScope SDK来下载数据集 #### 下载方法 :modelscope-code[]{type="sdk"} :modelscope-code[]{type="git"}ActionArt数据集的相关介绍刊载于论文《ActionArt：推进面向细粒度以人为中心的视频理解的多模态大模型（Multimodal Large Model）》（即将发表）

搜集汇总

背景与挑战

背景概述

ActionArt数据集源自相关研究论文，专为细粒度的以人为中心的视频理解设计，包含训练集和测试集。训练集提供视频名称和手动标注的字幕，测试集则分为多个子集，每个子集以JSON文件形式记录视频名称及对应的人工标注问答。该数据集基于Movid数据集构建，需登录后下载使用。