Home · Open data · Perception · Human video

HoloAssist

169 hours of HoloLens 2 video of people fixing and operating devices with a remote instructor, 2,221 sessions with gaze, hands, depth and IMU.

Published by Microsoft Research, 2023-10

Egocentric view of hands operating an espresso machine
Open data

Frame from the HoloAssist dataset, Microsoft Research, CDLA-Permissive 2.0.

Commercial use OK

CDLA-Permissive 2.0

PriceFree

The publisher's licence permits commercial use and redistribution, with attribution.

Get dataset Opens holoassist.github.io in a new tab
From a terminal
wget https://hl2data.z5.web.core.windows.net/holoassist-data-release/video_compress.tar

Per-modality tar files on Azure static hosting: video_pitch_shifted.tar 184.20 GB, video_compress.tar 144.62 GB, ahat_depth.tar 560.46 GB, hands.tar 219.24 GB, eyes.tar 2.45 GB, head.tar 4.67 GB, imu.tar 4.63 GB, cam_info.tar 10.07 GB, labels JSON 111 MB, splits zip on the project site. Buyers pick components. Paper reports 166 hours, site says 169.

Specification, as published

Hours (stated)
169 h
Episodes (stated)
2,221 episodes
Scale
169 hours of recorded data, 350 unique instructor-performer pairs
Task
Repair & maintenance, Assembly, Other
Environment
Indoor – office, Indoor – lab
Modality
RGB, Depth, Audio, IMU, Eye gaze, Hand pose, Language
Capture
Egocentric
In frame
Person hands
Captured in
North America
Embodiment
human
Frame rate
29.5 fps
Resolution
896x504
Formats
MP4, Images (JPEG/PNG), JSON, Other
Size
about 1.13 TB (sum of listed components)
Hosted on
Azure Storage

Figures are the publisher's own. Kinetic Blocks has not measured this dataset.

About this dataset

HoloAssist captures two-person task sessions: a performer wearing a HoloLens 2 works on physical objects such as a coffee machine, a printer or flat-pack furniture while an instructor watches the egocentric feed and gives spoken guidance. The headset streams RGB, AHAT depth, eye gaze, hand and head pose, IMU and audio, and the release adds action segments, conversation labels and mistake annotations. It targets assistant-style tasks like mistake detection and intervention prediction. Released under CDLA-Permissive-2.0.

Publisher's own task names: GoPro, DSLR camera, Nintendo Switch, Nespresso machine, Espresso machine, office printer, IKEA furniture assembly, laser scanner, motorcycle, circuit breaker, Navvis device.

What Kinetic Blocks did, and did not do

Indexed and linked from the publisher. Not hosted, verified or graded by Kinetic Blocks. The publisher's terms govern. The licence shown is the one stated on the publisher's page, linked above. Nothing on this page is a claim by Kinetic Blocks.

Want the verified ones too?

Request access to the marketplace, where every paid listing is verified against its files and graded before it goes on sale.