delivery_id stringclasses 7
values | category stringclasses 7
values | duration_sec float64 456 936 | h5 stringclasses 7
values | video_left stringclasses 7
values | video_right stringclasses 7
values |
|---|---|---|---|---|---|
4725b9f29042 | Sorting, Storage and Packing | 647.272 | episodes/4725b9f29042/4725b9f29042.h5 | episodes/4725b9f29042/camera_left.mp4 | episodes/4725b9f29042/camera_right.mp4 |
b8635207ec61 | Utensil Cleaning | 465.759 | episodes/b8635207ec61/b8635207ec61.h5 | episodes/b8635207ec61/camera_left.mp4 | episodes/b8635207ec61/camera_right.mp4 |
65e847eaad14 | Gardening, Farming and Feeding | 495.122 | episodes/65e847eaad14/65e847eaad14.h5 | episodes/65e847eaad14/camera_left.mp4 | episodes/65e847eaad14/camera_right.mp4 |
38c09611c7e7 | Food Preparation and Cooking | 935.578 | episodes/38c09611c7e7/38c09611c7e7.h5 | episodes/38c09611c7e7/camera_left.mp4 | episodes/38c09611c7e7/camera_right.mp4 |
6328d7b4e446 | Pet Care | 543.219 | episodes/6328d7b4e446/6328d7b4e446.h5 | episodes/6328d7b4e446/camera_left.mp4 | episodes/6328d7b4e446/camera_right.mp4 |
21614e478cc2 | Repair, Assembly and Home Maintenance | 540.63 | episodes/21614e478cc2/21614e478cc2.h5 | episodes/21614e478cc2/camera_left.mp4 | episodes/21614e478cc2/camera_right.mp4 |
e1ccf6acfb1b | Clothing and Bedding Organization | 455.713 | episodes/e1ccf6acfb1b/e1ccf6acfb1b.h5 | episodes/e1ccf6acfb1b/camera_left.mp4 | episodes/e1ccf6acfb1b/camera_right.mp4 |
Egocentric Hand Data Samples
7 episodes, approximately 68 minutes, with stereo video, hand and head poses, timestamps, and camera calibration.
Files
Each episode contains one H5 file and two original camera videos:
episodes/<delivery_id>/
├── <delivery_id>.h5
├── camera_left.mp4
└── camera_right.mp4
episodes.csv lists the episodes and categories. manifest.json provides file paths, sizes, and checksums.
Download
pip install huggingface_hub h5py
hf download Oceanveo/ego-hand-samples --repo-type dataset --local-dir ./ego-hand-samples
Read an episode
import json
from pathlib import Path
import h5py
root = Path("ego-hand-samples")
episode = json.loads((root / "manifest.json").read_text())["episodes"][0]
h5_path = root / episode["h5"]
with h5py.File(h5_path, "r") as f:
sync = f["sync/stereo_pair_v1"]
left_frame_ids = sync["left_frame_id"][:]
right_frame_ids = sync["right_frame_id"][:]
time_ms = sync["pair_relative_ms"][:]
hand_pose = sync["hand_pose"][:]
hand_valid = sync["hand_joint_valid"][:]
head_pose = sync["head_pose"][:]
head_valid = sync["head_valid"][:]
left_video = h5_path.with_name("camera_left.mp4")
right_video = h5_path.with_name("camera_right.mp4")
The frame IDs identify frames in the corresponding camera video. Use pair_relative_ms for sample timing and the validity masks when selecting poses. Sample intervals can vary.
- Downloads last month
- 185