Sample egocentric dataset
An illustrative overview of how a first-person household and commercial workflow dataset from Veda Solutions is organised.
Task categories
Household
Example tasks: folding clothes, cleaning surfaces, cooking, kitchen appliance handling, upholstery and furniture care.
Commercial
Example tasks: product handling, shelf stocking, packing, and workplace cleaning workflows.
Custom
Task taxonomies are defined with each client to match the behaviours their models need to learn.
Recording perspective and specifications
Footage is recorded from a first-person (egocentric) perspective using wearable-camera or smartphone setups, focused on the hands and the task area. Resolution, frame rate, and device setup are confirmed per project.
Annotations and metadata
Depending on scope, clips can include task labels, timestamps, environment metadata, and annotations such as object detection, segmentation, tracking, keypoints, or human activity labels.
Example file manifest
Illustrative example — structure and fields are agreed per project.
dataset/
├── manifest.csv # one row per clip
├── videos/
│ ├── HH_KITCHEN_0001.mp4
│ ├── HH_TEXTILE_0001.mp4
│ └── CM_RETAIL_0001.mp4
└── annotations/ # if annotation is in scope
└── HH_KITCHEN_0001.json
manifest.csv columns (example):
clip_id, category, task, environment, duration_s, recorded_on, qa_statusQuality checklist
Illustrative example of review criteria applied before delivery.
- Task performed matches the brief
- Hands and task area stay in frame
- No children, IDs, contact details, documents or private screens
- File name and metadata follow project convention
- Video plays end-to-end without corruption
- Reviewer sign-off recorded
Have a Data Collection or Annotation Project?
Whether you need real-world data collected, an existing dataset annotated, or an end-to-end data pipeline, Veda Solutions can build and scale the workflow around your requirements.