pengrui-rio/FEAGINE_HumanEgo_DataSample

Human Ego Hand Trajectory Dataset

This repository contains the first delivery batch, captured with six synchronized cameras at 1920×1080 resolution and 30 FPS.

Latest Release — v1.1 Bilingual Verified

Released on 2026-07-17. Status: verified. This is the current validated release with 15 tasks and 32 episodes. Episodes affected by corrupted video frames were excluded, and the remaining tasks were renumbered sequentially.

Overview

This package contains synchronized egocentric manipulation episodes recorded by six cameras. Each episode provides:

• six upright camera videos at 1920 x 1080 and 30 FPS;
• frame-aligned 2D hand observations for all available camera views;
• metric 21-joint 3D hand trajectories in one shared camera reference frame;
• bilingual dense action annotations with segment timestamps;
• a browser review page with a synchronized trajectory overlay and semantic timeline.

Only episodes that passed the delivery validation are included in this package.

Directory Layout

tasks/
  task_NNNN__english_task_name/
    episode_NNNN__episode_NNNNNN/
      episode_manifest.json
      videos/
        left_cam_left.mp4
        left_cam_right.mp4
        mid_cam_left.mp4
        mid_cam_right.mp4
        right_cam_left.mp4
        right_cam_right.mp4
      hand_pose/
        observations_2d.parquet
        trajectory_3d.parquet
        coordinate_system.json
      semantic/
        annotation.json
      timestamps/
        episode_timebase.json
      review/
        review.mp4
        annotation.html
calibration/
  headset_02_1080p.json
meta/
  episode_index.csv
  task_summary.csv
  delivery_summary.json
  index.html

Open meta/index.html to browse all episodes. Each episode also has a standalone review/annotation.html page. All links are package-relative and work after moving the dataset directory. The index and episode review pages provide EN, 中文, and Both display modes and remember the selected language in the browser.

3D Hand Trajectory

hand_pose/trajectory_3d.parquet contains one row per available hand and frame:

Field	Description
frame_index	Zero-based episode video frame index.
timestamp_s	Time in seconds relative to the episode start.
hand	left or right.
keypoints_3d	21 x 3 metric hand joint coordinates.
reference_camera	Always mid_cam_left in this package.

The coordinate unit is meters. The reference frame is the upright mid_cam_left optical frame:

• +X points to image right;
• +Y points to image down;
• +Z points forward from the camera.

The exact 21-joint order and axis definition are stored in hand_pose/coordinate_system.json.

2D Hand Observations

hand_pose/observations_2d.parquet contains frame-aligned observations from the six camera views:

Field	Description
frame_index	Zero-based episode frame index.
source_video_frame	Corresponding frame in the episode camera video.
timestamp_s	Time relative to the episode start.
camera	Camera name matching a file under videos/.
hand	left or right.
keypoints_2d	21 x 2 pixel coordinates in the upright 1920 x 1080 image.
confidence	Observation-level confidence when available.
keypoint_confidence	Per-joint confidence when available.
visibility	Visibility value when available.

Camera Calibration

calibration/headset_02_1080p.json contains the six camera intrinsics, OpenCV radial- tangential distortion coefficients, image dimensions, and the six extrinsic transforms.

For a 3D point p_ref from trajectory_3d.parquet, the calibration transform follows:

p_camera_h = T_reference_to_camera[camera] @ [p_ref.x, p_ref.y, p_ref.z, 1]

Project p_camera_h[:3] with the selected camera's camera_matrix and distortion_coefficients. Translation values and 3D trajectories are both in meters.

Time Alignment

All six camera videos in an episode share the same 30 FPS frame timeline. timestamps/episode_timebase.json records the episode-relative frame origin, duration, and source interval. The 2D observations, 3D trajectories, semantic segments, and review video use this same episode-relative timebase.

Semantic Annotation

semantic/annotation.json provides bilingual overall descriptions and temporally ordered action segments. Each segment includes start/end times, action labels, actor roles, interaction phase, contact state, and manipulated objects.

Review Media

review/review.mp4 is a 1920 x 720, 30 FPS H.264 visualization. It is encoded with a browser-compatible pixel format and fast-start metadata. It is a visualization asset; the authoritative numerical values are the Parquet files and calibration JSON described above.