HUMOTO
收藏资源简介:
HUMOTO数据集是由德克萨斯大学奥斯汀分校和Adobe Research联合创建的高保真人类对象交互4D动画数据集。该数据集包含736个动作序列,总计7875秒(30帧/秒),涵盖63种精确建模的物体和72个关节部件。数据集的特色在于详细的交互建模,特别是手部姿势的精确捕捉。数据集通过场景驱动的LLM脚本化管道创建完整、有目的的任务,并通过运动捕捉和相机记录设置处理遮挡。数据集适用于运动生成、机器人技术、计算机视觉和具身AI系统等多个研究领域。
The HUMOTO dataset is a high-fidelity 4D animation dataset for human-object interaction, jointly created by The University of Texas at Austin and Adobe Research. This dataset contains 736 motion sequences, totaling 7875 seconds (30 frames per second), covering 63 precisely modeled objects and 72 articulated parts. It features detailed interaction modeling, particularly the accurate capture of hand poses. The dataset generates complete, purposeful tasks via a scene-driven LLM scripted pipeline, and handles occlusions through motion capture and camera recording setups. It is applicable to multiple research fields including motion generation, robotics, computer vision, and embodied AI systems.

- 1HUMOTO: A 4D Dataset of Mocap Human Object Interactions德克萨斯大学奥斯汀分校, Adobe Research · 2025年



