Good episodes
Good episodes is for first-person recordings of real manual work. You divide the recording into episodes, judge each episode, and give a reason code for every unusable episode.
Watch before cutting
Watch the recording from start to finish in every synchronized view. Never mark an episode unusable from one view when another view may show the hand or the physical change.
If the entire recording is unreadable because the lens is covered, the image is black, or the file is broken, mark the whole recording unusable. If any part is useful or the venue can be identified, continue cutting and mark only the bad parts unusable.
Cut episodes
An episode is one continuous task with one clear goal. Several actions may belong to one episode; do not cut every action separately.
- An episode is at most five minutes.
- Cut at a natural break where one task ends, another begins, or a clear pause occurs.
- Never cut in the middle of an action.
- Repetition is allowed; keep annotating repeated instances of the same task.
- Every second must belong to exactly one episode. No gaps or overlaps.
Three usable criteria
All three must hold for an episode to be usable:
A Real-world value
The activity must create useful, visible value at work, home, or in the community. Usable examples include washing dishes, stocking shelves, assembling, cooking, folding or sorting laundry, mopping/wiping/sweeping, repair, watering, weeding, and planting.
Unusable examples include playing with toys or fidgeting, idle waiting or watching food cook, talking or gesturing without object interaction, sleeping or resting, random movement without a goal, eating or drinking when the key action is hidden, playing an instrument, and driving or navigating.
B Visible and active hand
At least one supplied view must clearly show at least one hand actively manipulating an object.
- Camera occlusion: check other views; one clear view is enough.
- Task occlusion: a hand naturally hidden by the task, such as under cloth or inside a bag, is expected and can remain usable.
- Unexplained loss: if every view loses the hand and the task does not explain it, mark
occluded.
C Visible physical change
The action must produce a change visible to the camera. Check every view before deciding that no change is visible.
Reason codes
Use the exact spelling and choose the best single code for each unusable episode:
| Code | Use when |
|---|---|
personal_activity | Phone use, eating, drinking, toilet use, sleeping, or resting |
idle | Standing, pausing, sitting, or waiting without meaningful activity |
in_transit | Empty-handed travel between areas, driving, or riding in a vehicle |
camera_not_worn | Camera placed on a table, held in a hand, in a bag, or in a pocket |
camera_setup | Putting on, removing, or readjusting the camera |
occluded | Hand, clothing, hair, or another object blocks the view or key action |
too_dark | Too dark to identify the task, hand, or objects |
too_bright | Overexposed so the task cannot be seen |
blurry | Out of focus, severe motion blur, or violent shake |
sensitive_content | Nudity, blood or injury, pornography, or explicit violence |
broken_video | Black screen, frozen image, or corrupted recording |
Walking, pauses, and phone use
- A pause or wait of 15 seconds or less can remain in a usable episode when the task stays continuous.
- A phone check of 5 seconds or less can remain when the task immediately resumes.
- Walking within the work area remains usable when the person is still doing the task.
- Carrying an object to another room remains part of the task when the hand is actively handling it.
- Empty-handed travel without a clear purpose is a separate
in_transitepisode. - More than 15 seconds idle, stopped work, or phone use over 5 seconds becomes a separate unusable episode with the relevant reason.
Final check
Confirm complete coverage, no cuts through an action, a reason on every unusable episode, and review of every supplied camera view.