Vision API - Scalpal

Detect

POST /detect

{
  "image": "<base64 JPEG, PNG or WebP>",
  "labels": ["scissors", "forceps", "gloved hand", "lap_scissors"],
  "threshold": 0.2
}

labels is optional and defaults to the instrument and hand set. Labels can be free text or catalog ids, up to 32.
threshold defaults to 0.2. Use 0.3 or more on cluttered scenes.
Images are downscaled so the long side is at most 1024 px. Encoded images over 8 MB are rejected.

{
  "detections": [
    {
      "label": "scissors", "score": 0.91,
      "box": { "x": 0.0, "y": 0.057, "w": 0.971, "h": 0.89 },
      "id": "lap_scissors", "kind": "instrument", "candidateIds": ["lap_scissors"]
    }
  ],
  "model": "owlv2", "ms": 458.0, "image": { "width": 278, "height": 480 }
}

Boxes are normalized 0 to 1 with the origin at the top left, in camera image coordinates. id is set when a label maps to exactly one catalog tool; otherwise check candidateIds. Errors return {"error": code}: invalid_request (422), invalid_base64, invalid_image, unsupported_image_format, empty_image and image_too_large (413).

Other routes

GET /health returns {status, model, device, warmupMs}.
GET /models returns the active model, the backends and the default labels.

​Detect

​Other routes

Detect

Other routes