Video Annotator

What the work is for

Labelled video teaches models to follow objects and recognise actions and events as they unfold over time.

The work

• Track objects through a clip. Set boxes on selected frames and correct the boxes the tool adds between them.
• Mark the frames where a tracked object is hidden or leaves the picture.
• Split clips into segments and label the action or event in each one.
• Raise clips the guideline does not cover instead of guessing.

What to expect

• Each project has its own guideline, label set and tool, for example CVAT run on YPAI's own servers.
• You pass initial test tasks before working on project batches.
• Every batch is reviewed. Clips that miss the acceptance criteria come back to you to fix.
• On some projects a tracking model proposes the boxes and you correct them.

What you bring

• A reliable internet connection.

You need

• Strong attention to detail, held from the first frame to the last.
• Comfortable learning a new web-based tool when a project needs one.
• 18 years or older.

Good to know

• Frame-by-frame work is repetitive and needs steady concentration.
• Work comes project by project, and there can be gaps between projects.