BOSS
收藏资源简介:
BOSS数据集是由新加坡-麻省理工学院研究与技术联盟创建的一个大型多模态视频数据集,专注于在物体-上下文场景中预测人类信念状态。该数据集包含900个视频,总计347,490帧,通过捕捉非言语沟通信号如目光、姿态和上下文信息来记录参与者的信念状态。数据集的创建过程涉及招募参与者进行特定的协作任务,并通过创新方法获取信念状态的精确标注。BOSS数据集旨在为研究机器理论的心智模型提供标准,特别是在需要非言语沟通的环境中,如人机交互和协作领域。
The BOSS Dataset is a large-scale multimodal video dataset developed by the Singapore-MIT Alliance for Research and Technology, focusing on predicting human belief states in object-contextual scenarios. This dataset includes 900 videos with a total of 347,490 frames, and records participants' belief states by capturing nonverbal communication cues such as eye gaze, body posture and contextual information. The creation process of this dataset involves recruiting participants to complete specific collaborative tasks, and acquiring precise annotations of belief states through innovative methodologies. The BOSS Dataset aims to serve as a benchmark for researching theory of mind models for machines, especially in scenarios requiring nonverbal communication such as human-computer interaction and collaborative fields.
- 1Revisiting Out-of-distribution Robustness in NLP: Benchmark, Analysis, and LLMs Evaluations · 2023年



