Egocentric video datasets, collected to your brief
Public egocentric video datasets are a good place to start research. When a model needs footage of specific activities, from specific places, with a licence that covers commercial training, the next step is to have it collected. GoKuna collects first-person video to a written brief, through contributors who have agreed to how it is used, and reviews every clip before it counts.
What an egocentric dataset is
An egocentric dataset is video recorded from the viewpoint of the person acting: a camera at head height or on the chest, looking where they look, with their hands and the objects they handle in view. It is the natural training data for models that need to understand how people carry out tasks, such as recognising activities, anticipating the next step or following hand and object interactions.
What makes one dataset more useful than another is consistency and coverage: the same camera position and clip structure across many people, many homes and many variations of the same activity.
The first-person footage GoKuna collects
Household tasks: short point-of-view clips of ordinary jobs around the home, such as cooking, tidying and laundry, filmed with a phone or action camera at head height. Contributors are recruited worldwide.
Long-form recording: longer stretches of everyday life from a head or chest mount, showing normal activity as it happens rather than staged scenes. This runs in North America to start.
Both run through the same task flow: a written brief, a reservation, a recording, a preview by the contributor and a review against the brief.
What you can specify
A brief turns your requirements into a task contributors can follow. It sets the capture instructions, whether the task is video or still images, the minimum and maximum clip duration, whether audio is required, the file formats and the largest file accepted, the number of submissions wanted and, where it matters, which countries contributors may be in.
Say which activities you need, where the camera sits, how long each clip runs and what must stay out of frame. The clearer the brief, the more consistent the footage.
Consent and privacy built into every clip
Contributors are adults who confirm they are 18 or older and agree to only submit media they are authorized to share. Each task records the consent terms that applied to it. The capture rules tell contributors to keep identifiable bystanders, private information and other people's property out of frame without permission, and contributors preview every clip before it is uploaded.
Contributors keep ownership of their recordings and grant a licence for training and evaluating machine learning models, including on behalf of GoKuna's customers. Your organisation is not named to contributors.
Review before anything counts
Every submission is reviewed by a person against the stated requirements before it is accepted. Clips that miss the brief are rejected with a reason, and the contributor can record again while the task is open, so the collection converges on what you asked for rather than on whatever was uploaded first.
Contributors are paid only for accepted submissions, which keeps the incentive pointed at following the brief.
Planning the shape of the collection
Decide first what varies and what stays fixed. Usually the camera position, the clip structure and the file format stay fixed, while the activities, the homes and the people vary. Write the fixed parts as instructions and the varying parts as the scope: which activities, how many clips of each, and from which regions.
If you are unsure of the numbers, start with a smaller first collection, review what comes back, and tighten the brief before scaling up. Every requirement you add is one more thing contributors follow and reviewers check, so add the ones your model actually depends on.
How a collection starts
Send a brief from the form on the homepage: your email and a description of what you need, how you will use it, and any budget or timing you have in mind. GoKuna replies by email with a quote, and scope and price are agreed before any collection starts.
A useful first brief can be short, for example the activities, the number of clips, the regions and the camera position. Details can be settled over email.
What is an egocentric dataset?
Video recorded from the viewpoint of the person doing the activity, with the camera at head height or on the chest, so their hands and the objects they use are in view.
Can I choose where contributors are?
Yes. A task can be limited to specific countries. Long-form first-person recording runs in North America to start.
Are the people in the footage identifiable?
The camera faces outward from the contributor, so their face is not in shot. Contributors are told to keep identifiable bystanders out of frame unless they have permission.
How is pricing set?
By quote. You describe the collection, GoKuna replies by email, and scope and price are agreed before anything is collected.