xuejf/EgoStandard-v2-Test

EgoStandard

9,000 hours of head-view egocentric activity across three progressive supervision configs.

Collection · EgoPro · EgoDemo · Interactive Visualizer · 中文

Private v2 test repository. This card, schema, and visual design are ready for review. Audited Parquet manifests, media, approved license, checksums, and final citation are not present, so the Dataset Viewer stays disabled and the loading examples below are release-target examples rather than currently executable downloads.

Hero SKU: EgoFull

EgoFull is the highest-supervision config in the Standard head-view series. It binds head video, 3D hand pose, 3D body pose, and V7 semantic references to one stable clip_id.

clip_id
|-- head_video_ref
|-- hand_pose_3d_ref
|-- body_pose_3d_ref       # EgoStand-motion and EgoFull
`-- semantics_v7_ref       # EgoFull only

Higher configs add references to aligned annotation layers. They do not imply that media is duplicated or that EgoStandard and EgoPro share every source clip.

Configs and hours

Config	Hours	Capture and annotation contract	Role
EgoStand	8,400	Head video + 3D hand pose	Base
EgoStand-motion	500	Base + 3D body pose	Motion
EgoFull	100	Motion + V7 semantic references	Hero SKU
	9,000		Repository total

Interactive environment and task dashboard

Open EgoSuite-Visualizer to explore the seven-environment distribution, hover-linked media panel, task ranking, and EgoFull versus EgoProMax mode switch.

Dataset Cards cannot run arbitrary JavaScript or reliably embed an iframe. The Space is therefore the executable interaction surface; this card keeps a static preview and a direct link.

Supplied task examples

Only six task names were supplied for this test iteration. They are shown without fabricated counts or ranks:

Task	Chinese
Replace Shoelaces	更换或系鞋带
Winding Charging Cable	缠绕收纳充电线
Gift Box Wrapping	礼品盒包装
Organize / Fold Clothes	衣物折叠与整理
Refill Tissue Box	纸巾盒补充
Seal Envelope	信封封口

The Visualizer reserves fourteen additional Top 20 slots as pending_audit until an audited task-frequency export is supplied.

Target loading interface

from datasets import load_dataset

egofull = load_dataset(
    "xuejf/EgoStandard-v2-Test",
    "EgoFull",
    split="train",
    streaming=True,
)

sample = next(iter(egofull))
print(sample["clip_id"])
print(sample["hand_pose_3d_ref"])
print(sample["body_pose_3d_ref"])
print(sample["semantics_v7_ref"])

Common manifest fields

Field	Type	Meaning
clip_id	string	Stable anonymous clip key
config_name	string	Exact Hugging Face config name
scene_family	string	One of the seven environment families
task_name	string	Normalized task name
duration_s	float32	Audited usable duration
head_video_ref	string	Head-view media reference
hand_pose_3d_ref	string	Frame-aligned hand-pose reference
body_pose_3d_ref	string, nullable	Body-pose reference
semantics_v7_ref	string, nullable	V7 semantic reference
release_revision	string	Immutable source revision
sha256	string	Primary media integrity checksum

Access, license, and citation

The approved release target is public discoverability with manual gated access. Access requests must not be automatically approved until a responsible reviewer, review criteria, and notification address are assigned.

The requested cc-by-nc-sa-4.0 identifier is not declared in this test card because no approved dataset license was supplied. Public visibility, access approval, and repository membership do not independently grant rights to unpublished data.

Users must not identify participants, reconstruct sensitive locations, or use the data for surveillance, profiling, or harmful applications.