Skip to main content Diffraction

The public collection · 2 releases

Real tasks.
Inspectable data.

Explore our egocentric dataset samples by task and signal. See what was captured, how it was processed, and what to evaluate before training.

Diffraction on Hugging Face

Get the files, explore dataset cards and follow our organization for new releases.

Explore on Hugging Face

Browse public dataset samples

Capture modality

2 of 2 releases

Actual appliance maintenance footage from Diffraction’s public egocentric sample RGB · Maintenance

Public sample · CC BY 4.0

Appliance maintenance

Tape removal, brushing and wiping from a chest-mounted view. Six excerpts with source-frame mapping and approximate action intervals.

6 curated excerpts176.84 s
Explore maintenance sample
Three kitchen recordings with RGB frames, metric depth and sensor confidence RGB-D · Kitchen

Public sample · CC BY 4.0

Kitchen manipulation

Onion preparation, pepper slicing and dicing. Three recordings with native RGB-D, device sensors and estimated annotations.

3 recordings167.85 s
Explore kitchen sample

Choose by the evidence you need

Different tasks. Different signals.

Appliance maintenance / RGB

Included signals
RGB, source-frame mapping, estimated hands and objects, 13 approximate action intervals.
Coverage
One wearer, one appliance/work area and one apparent session.
Limits to account for
No depth, metric camera calibration/pose, IMU, robot commands or execution-success labels.
Read the release notes ↗

Kitchen manipulation / RGB-D

Included signals
RGB, metric depth and confidence, calibration, device IMU, estimated camera pose, hands and objects.
Coverage
One conservative session group; participant count and independent session identity not supplied.
Limits to account for
No measured robot commands, force or tactile data. Raw IMU acceleration units require verification.
Read the release notes ↗

These small human-observation releases support technical inspection. Neither reports independent annotation accuracy or demonstrated robot-policy improvement. Their counts describe different capture units and should not be treated as independent trial counts.

From public sample to your collection

Build around your question.

Define the task, signals and acceptance criteria with our buyer’s guide, or bring your requirements directly to the team.

Read the guide & get the brief template →
Discuss a custom collection ↗