Each task is a set of frames with an end. Progress is shared: a frame anyone has answered counts as done.
Boxes are model guesses — fix them. Tap a box or a crop below to select · tap a class to change it · drag to move/resize · + Add box then drag on the image.
Tap the answer: it saves and the next frame opens. ◀ Back to change one. Tap the frame to zoom in where you tap; tap a zoom crop to enlarge it. Keys are on the buttons · s skip.
🎉 This task is finished for you — nothing left to answer here. Thank you!
History
Reviewer dashboard
Tasks — progress per task and who did it; switch a task off to hide it from labellers
All frames
Stream coverage
Reviewers — tap “Review” to sample a reviewer's labels at random
Review accuracy — from your Correct/Wrong calls on each reviewer's labels
Labels collected
Classifier progress — auto-retrains every ~500 labels; vs the VLM+detector ensemble baseline