Case study · Medical and clinical imaging
Gaze and activity labels for first-person clinical imagery
Frame-level labels for imagery from head-mounted clinical devices, under restricted-room conditions, for attention and activity models.
- Frames labelled
- More than 5.5 million
- Gaze-target agreement
- 97.8%
- Text and numeral transcription
- 99.6% exact
The work
The client supplies keyframes captured by head-mounted clinical devices during their wearers’ everyday activities, with the wearers’ consent. Our labellers mark what the wearer is attending to, the direction of attention of other people in the frame, posture, and any text or numerals in view.
Because the imagery is first-person and includes private places and bystanders, the project runs only in restricted rooms at our Nairobi and Manila centres, at content tier S3, with wellbeing time on the hour and redaction rules agreed with the client before work began.
The work is under a confidentiality agreement and the client is not named.