Voxel51/Egocentric-EMG-Force

Egocentric RGB-D + EMG/Force → FiftyOne (Native Multimodal MCAP)

The Lo6yu/egocentric_dataset daily activity release, converted to native multimodal MCAP episodes.

Someone wearing a depth camera and an EMG band on each wrist works through eight household tasks: sweeping and mopping, tidying a bedroom, a dining room and a living room, loading a washing machine, washing hands at a sink, wiping hands on paper towel, and applying hand cream. Alongside the video the release carries the muscle signal from both forearms, a hand skeleton per frame, and an estimate of how hard each finger is pressing.

Every episode is cut into subtasks a reviewer has described in English, so the 122 segments read as instructions rather than indices: pick up dustpan and broom, fit the trash bag into the basin, restock paper rolls.

Installation

pip install fiftyone

Usage

import fiftyone as fo
import fiftyone.utils.huggingface as fouh

dataset = fouh.load_from_hub(
    "Voxel51/Egocentric-EMG-Force",
    name="Egocentric-EMG-Force",
    persistent=True,
)
fo.launch_app(dataset)

The tasks with the firmest contact:

view = dataset.match({"peak_finger_force": {"$gt": 40}})

Episodes that involve folding:

view = dataset.match({"phases": {"$elemMatch": {"$regex": "fold"}}})

What you get

Eight episodes and 38.2 minutes of recording. Each episode carries:

• /camera, the first-person view at 640x480, as foxglove.CompressedVideo
• /depth, the matching Z16 depth frame, as 16-bit PNG
• /emg-left.plot and /emg-right.plot, eight channels per wrist at roughly 550 Hz
• /wrist-imu-left.plot and /wrist-imu-right.plot, accelerometer, gyro and orientation from each band
• /finger-force-{left,right} and /finger-force-{left,right}.plot, newtons for thumb, index, middle, ring and pinky
• /finger-contact-{left,right}.plot, the per-finger contact confidence
• /finger-flexion-{left,right}.plot, mean joint flexion per finger
• /hand-keypoints, the 21-point skeleton for each visible hand, as foxglove.SceneUpdate
• /subtask and /subtask.plot, each segment's English label at the frame it begins
• /instruction, the task the episode is carrying out

Across the whole set that comes to 67,185 camera and depth frames, 2,539,493 EMG samples, 60,254 hand skeletons, 133,721 per-finger force readings and 122 described subtask segments over 121 distinct phases.

Episodes carry the fields package_id, task, task_description, scene_description, subtasks, phases, objects, num_camera_frames, num_depth_frames, num_emg_samples, num_hand_keypoint_frames, num_force_frames, num_subtasks, has_emg, peak_finger_force, duration, fps, rgb_width and rgb_height.

Notes on the conversion

Video is re-encoded to Annex-B H.264 without B-frames. The source stores it as lossless FFV1 in MKV, which the viewer does not decode.

Depth is carried as 16-bit PNG at its published scale. The source stores raw Z16 frames in a zstd stream with no per-frame header.

The wrist EMG leads the camera by 80 ms. The release records that offset and publishes a corrected clock, which is the one used here, so the muscle signal lines up with the frame it belongs to.

Contact force is an estimate derived from hand pose and depth, not a reading from an instrumented glove. It saturates at 45 N; across the set only a handful of samples reach that ceiling.

Hand joints are published in the camera frame as the release computes them, and each skeleton is stamped from the camera frame it was measured on. A frame occasionally reports the same hand more than once, and the first detection of each side is the one published.

hand-cream-application names its contact columns dynamic_finger_force_N_* where the other seven write finger_contact_force_N_*. Both are read.

License & attribution

The source dataset is released under CC BY-NC 4.0, and this conversion is distributed under the same license. Use is limited to non-commercial purposes.

Changes from the source: conversion to the FiftyOne MCAP flavor, re-encoding of the video to H.264, decoding of the Z16 depth stream to PNG, and encoding of the sensor tables and semantic labels as message streams.