Egocentric Human Dataset Sample
数据集简介 | Dataset Overview
Egocentric Human Dataset Sample 是由 SeeWay.ai 整理的第一人称视角人类活动数据样本,包含视频、逐帧时间戳、加速度计、陀螺仪及采集记录。数据可供第一人称视觉理解、视频表征学习和具身智能相关研究探索;适用性取决于具体任务与数据质量。
This dataset, curated by SeeWay.ai, contains egocentric human-activity video samples with frame timestamps, accelerometer and gyroscope records, and capture metadata. It may support research in egocentric visual understanding, video representation learning and embodied AI, subject to the suitability of the data for each task.
数据概况 | Dataset Summary
以下为 2026 年 11 月 12 日对仓库内容的盘点;后续上传或修改可能改变统计结果。
The following is a repository snapshot taken on November 12, 2026. Counts may change as the dataset is updated.
| 项目 / Item | 内容 / Value |
|---|---|
| 场景类别 / Categories | 8 |
| 样本组 / Sample groups | 91 |
| 视频文件 / Video files | 91 MP4 |
| 仓库文件大小 / Repository file size | 约 25.1 GB / approximately 25.1 GB |
| 元数据记录的总时长 / Sum of metadata durations | 40,177 秒,约 11 小时 9 分 37 秒 |
| 惯性传感器标称采样率 / Nominal IMU rates | 元数据中标注为 100 Hz 或 200 Hz;以各组文件为准 / 100 Hz or 200 Hz as reported in metadata; see each sample |
总时长是各组 duration_seconds 字段的简单求和,尚未逐一核对视频实际播放时长。一个样本组在当前仓库中对应一个 MP4 文件;不应据此假定一个样本必然等于一个任务 Episode。
The total duration is the sum of the duration_seconds metadata fields, not a frame-by-frame verification of video playback duration. Each current sample group contains one MP4 file; this does not establish that each sample is one task episode.
场景与样本数 | Categories
| 场景 / Category | 仓库目录 / Repository directory | 样本数 / Samples |
|---|---|---|
| 订单拣选与分拣 / Order Picking & Sorting | 01_Order_Picking_Sorting |
6 |
| 打包与拆包 / Packaging & Unpacking | 02_Packaging_Unpacking |
1 |
| 食品与饮料制作 / Food & Beverage Preparation | 03_Food_Beverage_Preparation |
32 |
| 餐饮服务 / Catering Services | 04_Catering Services |
1 |
| 清洁与保洁 / Cleaning & Janitorial Services | 05_Cleaning_Janitorial |
41 |
| 组装与生产 / Assembly & Production | 06_Assembly_Production |
1 |
| 维修与维护 / Repair & Maintenance | 07_Repair_Maintenance |
2 |
| 办公与文档处理 / Office & Document Processing | 08_Office_Document_Processing |
7 |
场景名称依据仓库目录和元数据整理;具体视频中出现的动作、任务步骤与结果请以实际样本为准。
Category names are based on repository folders and metadata. Consult each sample for the actions, task steps and outcomes actually present.
文件结构 | File Organization
以 01_Order_Picking_Sorting/Sample1-1/ 为例:
01_Order_Picking_Sorting/
└── Sample1-1/
├── segment_000.mp4
├── segment_000_frames.csv
├── segment_000_gyro.csv
├── segment_000_accel.csv
├── segment_000_meta.json
└── verification_report.txt
| 文件 / File | 内容 / Contents |
|---|---|
segment_000.mp4 |
第一人称视角视频 / Egocentric video |
segment_000_frames.csv |
帧序号 frame_index 和时间戳字段 timestamp_ns / Frame indices and timestamps |
segment_000_gyro.csv |
陀螺仪时间戳及三轴 x, y, z 数值 / Gyroscope timestamps and three-axis values |
segment_000_accel.csv |
加速度计时间戳及三轴 x, y, z 数值 / Accelerometer timestamps and three-axis values |
segment_000_meta.json |
活动名称、时长、设备平台、标称采样率和同步方法等元数据 / Activity, duration, platform, nominal rates and synchronization metadata |
verification_report.txt |
数据提供方记录的检查项目与结果 / Reported checks and results |
使用说明与已知限制 | Usage Notes and Limitations
时间戳与同步:多数样本的元数据将同步方法标为
estimated_creation_time,少数标为measured。时间基准、单位及转换方式需要进一步确认,不能仅凭元数据中的偏移量或报告结果认定已精确同步。传感器数值:CSV 包含三轴数值,传感器数值的单位: 加速计m/s², 陀螺仪rad/s。
Timestamps and synchronization: Most metadata records list
estimated_creation_timeand a few listmeasured. A spot check found different numeric ranges in some frame and gyroscopetimestamp_nsfields. Clock origins, units and conversions need further validation; a recorded offset or report result does not by itself establish precise alignment.Sensor values: The CSV files contain three-axis values; units are m/s² (accelerometer) and rad/s (gyro) respectively.
获取数据 | Access
数据集目前设置为申请访问、人工审核。请先使用 Hugging Face 账号申请并获得访问权限,再下载文件。
Access requests are currently subject to manual approval. Sign in to Hugging Face, request access and wait for approval before downloading the files.
数据集 / Dataset: https://huggingface.co/datasets/seewayai/Egocentric-Human-Dataset-Sample
安装 Hugging Face CLI 并登录后,可下载数据集:
After installing the Hugging Face CLI and signing in, download the dataset with:
hf download seewayai/Egocentric-Human-Dataset-Sample --type dataset --local-dir ./egocentric-human-dataset-sample
许可与引用 | License and Attribution
本数据集标注的许可为 Creative Commons Attribution 4.0 International(CC BY 4.0)。使用和分享时,请遵守许可要求,包括适当署名、提供许可链接,并在适用情况下说明修改。
The dataset is marked as licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). Please provide appropriate attribution, a link to the license and an indication of changes where applicable.
- 许可全文 / License: https://creativecommons.org/licenses/by/4.0/
- 数据集 / Dataset: https://huggingface.co/datasets/seewayai/Egocentric-Human-Dataset-Sample
引用示例 / Example attribution:
SeeWay.ai, Egocentric Human Dataset Sample, https://huggingface.co/datasets/seewayai/Egocentric-Human-Dataset-Sample, licensed under CC BY 4.0. Changes: [describe changes, if any].
许可可能不涵盖其他适用权利,例如隐私、肖像或第三方权利。
The license may not cover other applicable rights, such as privacy, personality or third-party rights.
- Downloads last month
- 291