Mobile Egocentric Video Dataset Collection
Budget: ₹750 – ₹1,250 INR
I’m assembling a large-scale egocentric video dataset captured entirely with everyday mobile phones. These recordings will serve as raw material for training AI models focused on first-person perception, so realism and variety are essential.
Scope
• You will record short, first-person clips while performing ordinary activities—walking through shops, cooking, commuting, working at a desk, interacting with objects, etc.
• All footage must come from a chest- or head-mounted perspective using your phone (a simple lanyard, strap, or DIY mount is fine). Hand-held shots won’t meet the egocentric requirement.
• Natural lighting, ambient sound, and spontaneous motion are welcome; please avoid staged scenes.
• Each clip should run 1–3 minutes, saved in the phone’s native resolution and codec. No post-processing is needed.
Deliverables
1. A minimum of 60 unique video files (MP4 or MOV) totalling at least 90 minutes of footage.
2. A CSV log containing for every file: location type (kitchen, street, office, etc.), activity label, date/time, and device model.
Acceptance Criteria
• All videos start immediately from a first-person viewpoint without visible mounting gear or the recorder’s hands obscuring the lens.
• Footage is stable enough for computer-vision processing (minor shake is fine; excessive motion blur is not).
• No personally identifying faces or sensitive information appear; if unavoidable, blur them before submission.
• File naming matches the convention I’ll provide once we start.
Please briefly describe your phone model, the mounting method you can use, and examples of environments you can record in so I know the dataset will be suitably diverse.
Scope
• You will record short, first-person clips while performing ordinary activities—walking through shops, cooking, commuting, working at a desk, interacting with objects, etc.
• All footage must come from a chest- or head-mounted perspective using your phone (a simple lanyard, strap, or DIY mount is fine). Hand-held shots won’t meet the egocentric requirement.
• Natural lighting, ambient sound, and spontaneous motion are welcome; please avoid staged scenes.
• Each clip should run 1–3 minutes, saved in the phone’s native resolution and codec. No post-processing is needed.
Deliverables
1. A minimum of 60 unique video files (MP4 or MOV) totalling at least 90 minutes of footage.
2. A CSV log containing for every file: location type (kitchen, street, office, etc.), activity label, date/time, and device model.
Acceptance Criteria
• All videos start immediately from a first-person viewpoint without visible mounting gear or the recorder’s hands obscuring the lens.
• Footage is stable enough for computer-vision processing (minor shake is fine; excessive motion blur is not).
• No personally identifying faces or sensitive information appear; if unavoidable, blur them before submission.
• File naming matches the convention I’ll provide once we start.
Please briefly describe your phone model, the mounting method you can use, and examples of environments you can record in so I know the dataset will be suitably diverse.