TeleEgo
收藏资源简介:
TeleEgo是一个用于评估第一人称视角人工智能助手在现实世界场景下能力的长期、流式、全模态基准数据集。数据集包含来自多个参与者的同步视频、音频和文本数据,每个参与者贡献超过14小时的记录。所有数据流都精确地对齐到一个统一的全球时间线上,并丰富了手动编辑的语音转录和视觉叙述,以确保高质量和语义清晰。TeleEgo定义了12个诊断子任务,涵盖三个核心能力:记忆(回忆过去的事件)、理解(解释当前时刻)和跨记忆推理(将遥远的事件联系起来)。它包含3291个人工验证的问答项目,跨越多种问题格式(单选、二选、多选和开放式问题),在流式设置下严格评估。TeleEgo提供了对实际人工智能发展的现实和全面的评估。
TeleEgo is a long-duration, streaming, fully multimodal benchmark dataset for evaluating the capabilities of first-person perspective AI assistants in real-world scenarios. The dataset contains synchronized video, audio, and text data from multiple participants, with each participant contributing over 14 hours of recordings. All data streams are precisely aligned to a unified global timeline, and augmented with manually curated speech transcripts and visual narratives to ensure high quality and semantic clarity. TeleEgo defines 12 diagnostic subtasks covering three core capabilities: memory (recalling past events), understanding (interpreting the current moment), and cross-memory reasoning (connecting distant events). It includes 3,291 manually verified question-answering items spanning multiple question formats (single-choice, binary-choice, multiple-choice, and open-ended questions), which are strictly evaluated in streaming settings. TeleEgo provides a realistic and comprehensive assessment of real-world artificial intelligence development.



