The dataset viewer is not available for this split.
Error code: RowsPostProcessingError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Awaken-Human
English | 简体中文
Awaken-Human provides a recording-level preview from EngineAI-Ego, EngineAI's multimodal egocentric dataset for research in first-person perception, action understanding, and human motion analysis. Everyday activities are captured with a PICO headset in full-body tracking mode. The recording combines an undistorted left-eye video with body and hand tracking, camera calibration, and temporal action annotations.
The data supports investigation of first-person visual understanding, temporal action localization, video descriptions, human motion, and hand interactions. The sample enables researchers to inspect the data format and build data-loading workflows.
Current release
This repository currently distributes one complete sample recording. The sample configuration and sample split each refer to that same recording. The repository's Dataset Viewer is configured to preview its video and accompanying fields.
| Item | Sample |
|---|---|
| Source recordings | 1 |
| Temporal action segments | 8 |
| Video duration | 33.8 seconds |
| Video frames | 1,014 |
| Video encoding | H264, 512 × 384, 30 fps |
| Camera view | Undistorted left eye |
| Original files | Video, tracking, calibration, and action annotations |
Data structure
Awaken-Human/
├── README.md
├── README.zh-CN.md
├── LICENSE
├── LICENSE.zh-CN.md
├── SHA256SUMS.txt
├── data/
│ └── sample-00000-of-00001.parquet
└── processed/
└── <episode_id>/
├── trackingData_*.txt
├── CameraRecord_raw_undistorted_left_512.mp4
├── camera_params.json
└── action_labeling_<commit>_<timestamp>.json
The four recording files preserve their source filenames and the bytes stored in the source release. SHA256SUMS.txt lists checksums for the distributed files. The Parquet file provides one typed-video row and paths to the accompanying files; temporal segments remain part of the recording rather than separate dataset rows.
| File | Content |
|---|---|
trackingData_*.txt |
A tracking header with camera information, followed by timestamped JSON records containing head pose, body joints, and left/right hand joints when tracked. |
CameraRecord_raw_undistorted_left_512.mp4 |
Undistorted left-eye video at the source resolution and frame rate. |
camera_params.json |
Source camera calibration, including intrinsics, distortion parameters, and extrinsics. |
action_labeling_<commit>_<timestamp>.json |
Temporal segments with start_timestamp, end_timestamp, Chinese action_description, English action_description_en, and source action, hand, object, confidence, and noise fields. |
Timestamps in the action annotations use HH:MM:SS[.ffffff]. The Parquet row includes the recording identifier, video, relative file paths, duration, frame rate, frame count, image dimensions, segment count, and source segments. The original JSON is the recording's annotation file.
Usage
The repository is public and can be downloaded directly under the license below. Install datasets and huggingface_hub, then load the preview and download its recording files:
import json
from pathlib import Path
from datasets import Video, load_dataset
from huggingface_hub import snapshot_download
repo_id = "EngineAI-Embodied/Awaken-Human"
dataset = load_dataset(repo_id, "sample", split="sample")
row = dataset.cast_column("video", Video(decode=False))[0]
print(row["episode_id"], row["duration_seconds"], row["num_segments"])
root = Path(snapshot_download(
repo_id=repo_id,
repo_type="dataset",
allow_patterns=["processed/**", "LICENSE", "LICENSE.zh-CN.md"],
))
video_path = root / row["video_path"]
camera = json.loads((root / row["camera_path"]).read_text(encoding="utf-8"))
annotation = json.loads((root / row["annotation_path"]).read_text(encoding="utf-8"))
segments = annotation if isinstance(annotation, list) else annotation["segments"]
for segment in segments:
print(segment["start_timestamp"], segment["end_timestamp"],
segment.get("action_description_en") or segment.get("action_description"))
Video(decode=False) exposes the video reference without requiring video decoding during metadata inspection. Open the downloaded MP4 with a video player or use the video decoder appropriate for your environment.
License
This EngineAI-Ego sample is distributed under the EngineAI Academic Research Dataset License, version 1.0, for non-commercial academic research only. Commercial use and redistribution of all or any part of the data are prohibited, including re-uploading, mirroring, and distributing modified copies. Local preprocessing, annotation, and model training are permitted only for non-commercial academic research. Collaborators must obtain their own copies from the official repository.
Downloading or using the data constitutes acceptance of the license. Share the official dataset page instead of redistributing data. The English LICENSE is authoritative; LICENSE.zh-CN.md is its Simplified Chinese reference translation.
Citation
If you use Awaken-Human in academic research, acknowledge EngineAI and EngineAI-Ego, cite the official repository, and record the dataset revision used in your experiments.
@misc{engineai_awaken_human,
author = {{EngineAI}},
title = {{Awaken-Human}},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/EngineAI-Embodied/Awaken-Human},
howpublished = {Dataset}
}
- Downloads last month
- 4