First-Person Bench Work Video, Brazil (0.5 h starter set)
Verified supplier, South America
KB Quality grade is Kinetic Blocks' own grade, A+ to F, measured from the files across four pillars: Technical, Temporal integrity, Content, Completeness.
KB Quality grade is Kinetic Blocks' own grade, A+ to F, measured from the files across four pillars: Technical, Temporal integrity, Content, Completeness. How grading works
What you get
Measured hours
0.5 h
Measured by Kinetic Blocks from the files, not declared by the seller.
Clips
11 clips
KB Quality grade
A
Task
Other
Modality
Tabular, Text, 2D RGB video
Region
South America
Capture
Egocentric (camera worn by the person, first-person view)
Declared by the seller. Kinetic Blocks does not yet check the viewpoint against the files; open the free sample to see it.
File formats
json, csv, parquet, Other
Licence
Commercial Standard
Commercial use permitted, across your organisation. You may not redistribute, resell, or publish the dataset. Read more
Sale mode
Non-exclusive, can be sold to multiple buyers
Plain-English summary. The Data Licence Agreement you accept governs. Read the full terms
You pay Kinetic Blocks, never the seller. You get 5 business days to check the files against the dataset card. Reject for a material mismatch and, once we uphold it, you are refunded in full. How payment works
What we checked against the files
4 of 4 claims verified against the files
4 verified
0 can't confirm
0 doesn't match
Dataset card check
✓verified
Episode count check
✓verified
Episode Index
✓verified
Encoder
✓verified
KB Quality grade A, by pillar
A+
A
A-
B+
B
B-
C+
C
C-
D
F
Technical quality30%
93
Temporal integrity25%
100
Content usefulness30%
84
Completeness15%
100
Dead time is mapped, not cut.
100% active task footage (0.5 h of 0.5 h)
Violet is task action, grey is idle, measured by Kinetic Blocks in every clip. Idle stretches lower the grade; they are never cut from the files.
Scores out of 100. Every seller is graded on the same versioned weights. How grading works
0.5 h of head-mounted phone video of bench work (painting objects, nail care), hands in frame, 11 episodes.
Overview
First-person video of bench work recorded in Brazil on a head-mounted phone (ultra-wide lens, landscape, 1920x1080): painting objects and nail care, 3 recordings split into 11 task episodes, 0.48 hours, both hands visible 92.8% of the time.
Delivered as one LeRobotDataset v3 root, H.264 at an exact 30 fps and without audio, with the package metadata next to it: JSONL manifests, measured hand visibility per episode, self-calibrated camera intrinsics, AI-generated task steps (117 steps, 19 validated by people), a consent ledger and the accepted terms per recording.
This is a starter set from our ongoing collection in Brazil. More hours, other trades and environments, and a specific device mix are available as a commission, recorded to your specification.
Dataset card
# First-Person Bench Work Video, Brazil (Trades)
Head-mounted phone video of bench work in Brazil (painting objects, nail care), with the wearer's hands in frame.
## At a glance
| | trades |
|---|---|
| Recordings | 3 |
| Episodes | 11 |
| Hours | 0.48 |
| People | 1 |
| Tasks | 2 |
## Capture
- Phone on a head mount, ultra-wide lens requested, landscape. 1920x1080 for every episode.
- One phone model: Samsung SM-G990E.
- No audio: no video file in this dataset has an audio track.
- Environments declared by the person recording: salon and workshop bench.
## Hands
Measured on one frame per second of everything delivered.
- At least one hand visible 97.9% of the time, both hands 92.8%. Every recording has one or more hands in at least 90% of its time.
- Per episode: `hands_any_pct` and `hands_two_pct` in the episode manifest (`trades.lerobot/package/metadata/episodes.jsonl`).
## Annotations
- Task label per recording.
- Task steps: 117 steps in 11 of 11 episodes, each with start, end, verb, object, acting hand and one sentence in English and in Portuguese. AI-generated and marked `annotation_source: ai_generated`; 19 validated by people (`human_validated: true`).
- Camera intrinsics for all 3 recordings, self-calibrated (`intrinsics_origin`). None measured with a calibration target.
## Formats and folder structure
```
trades.lerobot/
<h264 root>/ LeRobotDataset v3 root: meta/, data/, videos/ (11 episodes, 11 video files)
derived/ frame maps of re-sampled episodes
package/ manifests, QA report, consent ledger, terms, Croissant card
index.json episode_index to episode_id, byte copy or derived
```
- The LeRobot root is the only folder in `trades.lerobot/` whose name ends in `-h264`; its name gives resolution, frame rate and codec.
- One LeRobot v3 root, 11 episodes in 11 video files. Camera key `observation.images.egocentric`. Tables carry `timestamp`, `frame_index`, `episode_index`, `index`, `task_index`. This is human video: no `action` or `observation.state` column.
- Every episode is H.264 at exactly 30 fps. 3 episodes are byte copies of the phone file; 8 recorded at a variable or non-integer rate were resampled to 30 fps and re-encoded in H.264 (CRF 18). `index.json` marks each one and `derived/<episode_id>.frames.csv.gz` maps every frame to its source frame.
- Episode length from 61.4 s to 180 s.
- Croissant 1.1 card with the Responsible AI extension in `trades.lerobot/package/croissant.json`.
## Loading
```python
from pathlib import Path
from lerobot.datasets import LeRobotDataset
root = next(Path("trades.lerobot").glob("*-h264"))
ds = LeRobotDataset("local/trades", root=root)
```
## Not included
No audio, IMU, depth, camera pose or 3D hand pose. No robot actions.
## Rights and consent
- First-party collection. The person recording is an adult who accepted written participation terms covering recording, AI training, commercial licensing and sublicensing.
- Each recording links to the terms version and the SHA-256 of the accepted text in `trades.lerobot/package/consent_ledger.csv`; the terms text is in `trades.lerobot/package/terms/`, with an English courtesy translation.
- Before delivery a person watched every recording to the end and confirmed there is no other person's face, reflection, document or screen with personal data.
- Participant ids are pseudonymous and valid only inside this dataset.
## License
Kinetic Blocks Commercial Standard licence, non-exclusive. The Kinetic Blocks Data Licence Agreement governs.
## Available per order
This is a starter set from our ongoing collection in Brazil. More hours, other trades and environments, and a specific device mix are available as a commission, recorded to your specification.