Skip to main content

Label

Source: app/api/label.py

POST /api/v1/label​

Start a SAM 3 auto-labeling job on one or more videos.

{
  "project_id": "...",
  "video_ids": ["..."],
  "prompts": ["person", "car"],
  "threshold": 0.5,
  "resolution": 1008,
  "fps": 5
}

Returns 202 Accepted with { job_id }. Poll /api/v1/status/{job_id} or subscribe to the WebSocket for progress.

POST /api/v1/label/exemplar​

Visual-prompt variant — instead of text, provide an image crop (or bbox) and SAM 3 finds visually similar objects.

POST /api/v1/label/preview​

Run SAM 3 on a small sample of frames to preview a prompt before committing to a full job. Returns base64-encoded JPEG previews.

POST /api/v1/label/segment-points​

Interactive segmentation. Send click points + frame and SAM 3 returns a mask. Used by the click-to-segment tool in the review canvas.

POST /api/v1/annotations​

Create an annotation manually (e.g. when a human draws a box in the review UI).